BUILT BY ONE OPERATOR.
4,254,093,040 ROWS.

Real voice fidelity, not generic AI: Will's full-corpus Professional Voice Clone (ElevenLabs PVC tier, voice_id vaqWG6OoW1Bb1g4nwV2H) drives both a portfolio Curator (visitor Q&A in the browser) and a Sales caller (Mac Septic phone tools). Same Pipecat pipeline, two personas, three transports — and the portfolio itself ships a 1-click voice CTA so prospects hear the agent before they read a single bullet.

Phil is a production voice agent built on Pipecat with a Professional Voice Clone of Will (ElevenLabs PVC tier, voice_id vaqWG6OoW1Bb1g4nwV2H, model eleven_multilingual_v2, stability 0.4 / similarity_boost 0.95 / style 0.7 / optimize_streaming_latency 2), Deepgram nova-2 STT, Silero VAD (curator tuned to confidence 0.8, start 0.5s, stop 0.5s, min_volume 0.15), and two Anthropic models: Haiku 4.5 for the Curator persona (fast portfolio Q&A) and Sonnet 4.6 for the Sales persona (CRM tool-reasoning). End-to-end latency runs ~1.4-1.8s with ~400ms on utterance end.

Multi-persona architecture: both personas share a `shared.build_pipeline(transport, persona)` factory — identical Deepgram + Anthropic + ElevenLabs + VAD stack, only the transport and persona system prompt differ. The Curator (portfolio.ecbtx.com) answers across the 39-project corpus with no tools, plus Phil-initiated lead capture (`[[ASK_EMAIL]]` marker → inline email form → Vercel function → CRM webhook) and a visitor-side mic picker (input device persisted in localStorage, handed to LiveKit `setMicrophoneEnabled({deviceId})`). The Sales persona drives Mac Septic phone calls with 8 CRM tools: book_appointment, check_availability, transfer_call, send_followup_sms, set_disposition, create_callback, leave_voicemail, end_call.

Three transports cover every entry point: LiveKit for the browser curator (with a same-origin `/api/livekit-token` proxy that bypassed the cross-origin CORS issue), Twilio Media Streams for outbound sales and the curator's `dial-curator` callback option, and RingCentral for inbound Mac Septic calls. Deepgram is pinned to nova-2 because the Pipecat deepgram-sdk pin throws WS-init 400s against nova-3; the live_options→settings migration is shipped.

• > Professional Voice Clone (Will) via ElevenLabs PVC tier
• > Two personas: curator (39-product portfolio Q&A) + sales (8 CRM tools)
• > Three transports: LiveKit browser, Twilio outbound, RingCentral inbound
• > Shared shared.build_pipeline factory across personas
• > ~1,400-1,800ms end-to-end latency, ~400ms utterance-end

• → SMS-during-call lead capture for outbound sales
• → Curator → Sales hand-off: detect sales intent and transfer the live session to the sales persona
• → Real-time transcript dashboard for in-call monitoring
• → Voice eval harness: replay synthetic visitor turns through both personas and grade responses

Eliminates lead-loss via cross-channel scoring; powers the entire Mac Septic operation; verticalizable to any field-service business.

ReactCRM is the operational backbone of Mac Septic. It unifies inbound calls, SMS, email, web chat, and browse activity into a single per-customer timeline with AI scoring. The cross-channel analyzer (shipped 2026-04-29) routes hot leads to a prioritized inside-sales queue and auto-suppresses unsubscribes. Browse-abandonment win-back (shipped 2026-05-08) recaptures permit-owner visitors with first-party tracking. Twilio caller-ID smart routing (629/931/737) selects the local number for each market, and PowerDialer notes sync writes call outcomes back to the lead record in real time.

The stack is React 19 on the frontend, FastAPI + PostgreSQL 16 on the backend, with Tailscale-mesh access to the on-prem permit database. It is built feature-flag-first so individual modules (scoring, dialer, abandonment) can be licensed independently to a second vertical.

• > 9,502 customer records
• > ~3K monthly email opens
• > ~500 daily chat inquiries
• > 200+ weekly calls

• → Productize as multi-tenant SaaS
• → Onboard 2nd vertical (plumbing or HVAC)
• → External API access tier

Septic clusters geographically (307 customers in a single Summit Glory Trail subdivision). The new rural_septic_score from FCC BDC absence pre-ranks the 107K unenriched candidates at zero cost. BatchData becomes the validation step on the top 10K, not the gating step on all 42K — turns a $4,200 pilot decision into a $1,000 validation pass.

The expansion engine joins TX OSSF (On-Site Sewage Facility) permits with subdivision polygons across 8 counties (Travis, Williamson, Bastrop, Hays, Comal, Bexar, Guadalupe, Kendall) to identify dense septic clusters. 690 high-confidence septic streets were extracted from the existing Mac Septic customer base as ground truth.

A pilot of $4,200 BatchData spend on Cedar Creek + East Austin (42,539 addresses) is projected to yield 150-200 net-new MAC customers, based on the historical cluster density observed at Summit Glory Trail (307 Mac customers in one subdivision). Enrichment is the gating step: 107,562 candidates remain unenriched and will be processed in cost-managed batches.

The 2026-05-16 acquisition of the FCC Broadband Data Collection (924M records, 116M unique US addresses) added a free pre-scoring layer. Every TX OSSF permit candidate now carries a rural_septic_score 0-100 derived from broadband absence + TIGER urban-area overlap + ZCTA population density. Mac Septic's 8,630 customer addresses validate the v2 model — they hit the 70+ score band at 4.19% vs 1.84% for a random TX baseline (2.29x lift). BatchData skip-trace is now reserved for confirmation on the top-ranked 4,190 addresses, not blanket enrichment of all 42K.

• > 165K TX OSSF permits in 8 counties
• > 107,562 unenriched candidates
• > 690 high-density septic streets
• > Summit Glory Trail = 307 Mac customers in one subdivision

• → Run rural_septic_score on the 107K unenriched candidates
• → BatchData confirmation on top-10K (cost cut from $4,200 → ~$1,000)
• → Feed pre-scored leads to Mac dialer

Roofers pay $50–100/lead for pre-qualified hail-strike addresses with recent roof permits; the data layer is already built and validated.

A 17.3M-row materialized view joins NOAA NCEI storm reports against permit data across TX. Validation cases: Tarrant Co May 27 2024 hail (3.25") drove 1,006 roof permits in the following 120 days; Bexar Co Mar 31 2025 hail (2") drove 3,406 permits. The product surfaces address-level lists pre-filtered by storm severity and roof age.

The data layers (austin_311, storm_events, spc_storm_reports, hail_leads MV) are live; the remaining work is a paid `/api/v1/hail_leads/search` endpoint with Stripe gating and a refresh cron. Real-world launch is imminent — the lead surface is live in the permitlookup-web Next.js app at hail.ecbtx.com, gated and ready for first pilot traffic.

• > 15,085 TX hail events since 2014
• > 862K unique hail-qualified addresses
• > 17.3M storm × permit rows
• > Demo export: 79K Fort Worth address-storm rows

• → Build /api/v1/hail_leads/search endpoint
• → Stripe gating
• → Sign 1–2 TX roofing partners

The largest data asset in the portfolio: ~2.5 billion rows of permits, property, business entities, code violations, mortgage data, and healthcare provider records across a primary + secondary Postgres cluster. Powers every permit-triggered sales motion in the portfolio and the moat for everything else.

PermitLookup is a nationwide records platform spread across a two-node Postgres cluster. The primary lives on a 32 GB RAM / 21 TB ZFS storage T430 and holds 1,426,233,456 rows across 176 tables in a 617 GB on-disk database. The secondary lives on an 88-core / 768 GB R730-2 and holds another 1,058,859,584 rows across 141 tables in 462 GB on disk. Combined, the cluster is ~2.5 billion rows and ~1.1 TB.

Permit coverage spans 57 state and territory tables on each node (all 50 states plus DC, Puerto Rico, Guam, American Samoa, Northern Mariana Islands, US Virgin Islands), totaling roughly 1.5 billion permit-row slots between the two with overlapping ingestion streams. The largest individual tables include Florida permits (230M on each node), code violations (297M on primary, 79M on secondary), Louisiana permits (99M on each), Texas permits (78M on each), business entities (99M / 35M), and New York permits (59M on each).

The cluster also carries property sales (97M primary, plus liens 37M, NYC ACRIS 33M, valuations 9M), HMDA mortgages (17M), septic systems (10M), national doctors and clinicians (12M), prospect contacts (15M), HMDA mortgages, food inspections, propane facilities, USTs, FAA facilities, sanctions entities, and a global census/demographics layer. State partitioning + a hot/cold split keeps query latency under 2 seconds even at the largest tables.

The API is FastAPI running on the R730 (88 cores / 768 GB) behind a Cloudflare Tunnel at permits.ecbtx.com, with a Railway-hosted backup at permit-api-production-6eae.up.railway.app and an `api_users` + `teams` schema already in place for multi-tenant self-serve. Licensing is the monetization path: H-Man Electrical is live, Crown Hardware is quoted, and a 5-contractor waitlist sits behind a self-serve checkout queued for shipment.

• > 2,485,093,040 rows across a 2-node Postgres cluster
• > 1.1 TB combined on-disk database footprint
• > Primary T430 (32 GB / 21 TB ZFS): 1.43B rows across 176 tables, 617 GB
• > Secondary R730-2 (88-core / 768 GB): 1.06B rows across 141 tables, 462 GB
• > ~1.5B permit rows across 57 state and territory tables per node
• > 50 US states + DC + PR + GU + AS + MP + VI covered
• > Layers beyond permits: code violations, property sales / liens / valuations, NYC ACRIS, HMDA mortgages, business entities, doctors and clinicians, food inspections, FAA facilities, USTs, sanctions, prospect contacts
• > Multi-tenant schema (api_users + teams) ready for self-serve
• > permitlookup-web Next.js 16 + shadcn frontend hosts dumpster-leads and hail-leads product surfaces on Vercel

• → Close 1–2 contractor licensing deals
• → Ship self-serve checkout (api_users + teams already provisioned)
• → External API access tier with usage-based pricing
• → Cross-node dedupe pass to publish a unified row count

Rural service businesses (septic, well-drilling, propane, satellite TV, off-grid solar) need to find addresses without sewer/fiber/cable. After five iterations, the v5 model delivers AI-validated lift: Mac Septic's customers score 5.9x higher in the high-rural bucket than HGAC Houston-area permits — apples-to-apples after symmetric spatial enrichment with 1.75M parcel polygons. Logistic regression independently confirms (AUC 0.86, lift 4.09x). 4,145 rural-Mac-style targets ranked and ready for Mac's dialer queue. The same engine generalizes to any rural-service vertical by swapping the validation cohort.

Rural Service Lead Engine v5 is a scoring layer over a 4.25-billion-row federal-and-state cluster: 924M FCC broadband records, 1.75M parcel polygons (HCAD 1.46M Harris + Comal 102K + Hays 119K + Bandera 32K + Kerr 34K = full Texas Hill Country + Houston coverage), 2,644 TIGER urban-area polygons, 33K census ZCTAs, and 228K TX OSSF septic permits. For every address it computes a rural_septic_score 0-100 combining: TIGER urban-area NOT-overlap (anti-signal -35), polygon-derived lot_acres (primary signal up to +35), population density, broadband absence, and joint-condition interactions.

The validation story: v1 used broadband absence alone (0.29x lift — failed). v2 added urban-area + density (2.29x lift — usable). v3 tried address-based parcel join (1.56x — regressed due to 0.012% match rate). v4 used HCAD spatial join (0.43x — inverted because Mac is in Hill Country, not Harris). v5 added Hill Country parcel polygons for symmetric enrichment of both cohorts (5.99x vs all-TX, 5.91x vs HGAC — SHIP). Logistic regression AUC 0.86, lift 4.09x independently confirms — no leakage.

Output: 4,145 Mac-customer-confirmed rural targets in the dialer-ready queue, plus 10,000 ranked TX OSSF permits as expansion candidates. Symmetric spatial enrichment is the key — both cohorts get apples-to-apples treatment so the lift number is real, not an artifact.

• > 5.9–6.2x lift vs HGAC baseline (v5 validation runs, 2026-05-17)
• > Mac parcel coverage 77.8% (8,630 customers → 6,718 with derived lot_acres)
• > 3.66M parcel polygons across 11 CADs: HCAD, DCAD, TAD, TravisCAD, WilliamsonCAD, HaysCAD, CCAD, BastropCAD, BurnetCAD, KerrCAD, BanderaCAD
• > 4,145 Mac-customer-confirmed rural targets ready for dialer
• > 10,000 ranked TX OSSF expansion candidates exported
• > 924M FCC broadband location-provider-technology records
• > Logistic regression: AUC 0.86, precision-lift 4.09x
• > 228,255 TX OSSF permits scored

• → FL + NC multi-state v5 (FL DOR 10.8M polygons + NC OneMap 5.72M polygons loading now)
• → Census-precision geocoding deployed (85% address-precision)
• → Outcomes feedback loop live in Mac CRM (conversion-by-score-band view, ML retraining path)
• → Real-time /v1/rural-score/lookup endpoint live for any US address
• → Sign first external rural-vertical pilot (well drilling, propane) using same engine

• ! HaysCAD polygons are coarse (median 653 acres) — subdivision-aggregate geometry inflates per-lot numbers, though the score formula bins at 10+ acres so it doesn't over-credit
• ! 27 Bandera customers share a single 67.9-acre polygon (subdivision-level geometry) — cosmetic; signal still says "rural Hill Country"
• ! Per-address geocoding (Mapbox/Census) would push Mac coverage from current 77.8% toward 95%+ at ~$0.005/address

Roofers pay $50-100 per pre-qualified storm-strike lead. The data layer is built: 15K TX hail events × 35M mortgage records × 97M property sales × 297M code violations. Two endpoints live: search by specific hail event, or pull last-14-days storm-strike candidates by state. Composite score = storm severity + home age + mortgage activity − recent roof-permit penalty.

`/v1/roofer-leads/by-hail-event?event_id=X&days_after=120&radius_miles=20&min_magnitude=1.0` returns properties within a hail event's footprint, ranked by composite score: storm_severity (hail magnitude × 10, capped 30) + home_age (years / 2, capped 25) + mortgage_score (+20 if HMDA match within window) − roof_recency_penalty (−20 if a roof permit was pulled <5y ago, −10 if <10y) × distance_falloff.

`/v1/roofer-leads/recent?state=TX&days_back=14&min_score=50&limit=100` runs the same scoring against all hail events from the last 14 days in a state.

The underlying data was acquired and indexed during the 2026-05-15 federal-bulk session: 771K NOAA storm events 2014-2025 (771,183 details + 690K locations + 11K fatalities), 35M HMDA mortgage applications 2020-2024 (every covered lender, every loan purpose, every action), 97M property sales, 297M code violations. The state-partitioned permit tables filter recent roof permits per state.

Delivery: free for existing PermitLookup customers; standalone pilot at $50-100/lead × 100 leads/mo per pilot customer.

As of 2026-05-17, Storm Strike Dispatch ships as a focused lead-acquisition product at storms.ecbtx.com. Buyers who want a full operating system (CRM + dispatch + estimates + claims) are routed to its sibling product RoofCRM at roofers.ecbtx.com — same data layer, different value prop.

• > 15,085 TX hail events since 2014
• > 771,183 NOAA storm events 2014-2025 (all event types)
• > 35M HMDA mortgage records 2020-2024
• > 97M property sales with year_built signal
• > 297M code violations as distress overlay
• > 2 production endpoints live on Railway + permits.ecbtx.com

• → Build roofers.ecbtx.com or /roofers landing with live lookup widget
• → Build a TX hail-event MV refreshed nightly for sub-second `/recent` response
• → Sign first TX roofer pilot (HOU/DFW preferred — highest event density)
• → Add insurance-claim-status overlay (FL DOI + TX DOI public adjuster filings)
• → Cross-sell to PermitLookup customers as a tier upgrade

• ! No active TX hail events in last 14 days at time of validation — `/recent` returned 0; benchmark on historical event needed
• ! Score weights are heuristic — needs first-customer feedback for tuning
• ! Distance falloff is geodesic, not actual storm-track polygon — over-includes addresses outside hail footprint

Roofer-vertical SaaS that bundles a CRM with the data layer behind it. Where ProRoof/Roofr/JobNimbus sell software, RoofCRM bundles software + Storm Strike Dispatch leads + claim history + permit context + parcel data. The differentiating thesis: a vertical CRM that already knows every TX roof permit pulled, every hail strike, every code violation, every HMDA mortgage. Same Crown Hardware / H-Man Electrical verticalization playbook applied to roofing.

RoofCRM is the operating system for Texas roofing contractors. The marketing site at roofers.ecbtx.com pitches a 6-feature product: AI-assisted estimation from drone photos, claim-status auto-sync from TX DOI public adjuster filings, storm-driven dispatch auto-queueing leads from the /v1/roofer-leads/* API, GPS-stamped photo documentation, a homeowner customer portal for job tracking, and Samsara-style crew/truck GPS.

The pricing layer maps to fleet size: Starter $400/mo per crew (1-3 trucks), Growth $800/mo per crew (4-10 trucks), Enterprise $2,000/mo per crew (10+ trucks). Differentiation vs ProRoof/Roofr/JobNimbus is built around the data-layer ownership story: RoofCRM is the only roofer CRM that comes with the underlying data (15K TX hail events, 35M HMDA mortgages, 97M property sales, 297M code violations, 1.75M parcel polygons across 11 CADs) bundled as a default feature — competitors integrate with separate paid leads providers.

Visually distinct from Storm Strike Dispatch despite sharing the underlying API: storms.ecbtx.com uses an urgent storm-warning orange-radar identity; roofers.ecbtx.com uses a professional copper/bronze blueprint identity (data-theme="roofers-os"). The split lets the same backend serve two buyer personas — Storm Strike for one-off lead packs, RoofCRM for full vertical operating-system subscriptions.

• > 6-feature product surface live at roofers.ecbtx.com
• > Distinct visual identity (copper/bronze blueprint, separate from Storm Strike orange-radar)
• > Integrated with /v1/roofer-leads/by-hail-event + /v1/roofer-leads/recent + permits.ecbtx.com
• > 3-tier per-crew pricing ($400 / $800 / $2,000/mo)
• > Comparison table vs ProRoof/Roofr/JobNimbus (data-layer ownership = the differentiator)
• > Cloudflare DNS + Vercel deploy on Next.js 16

• → Sign first pilot — reach out via existing H-Man Electrical and Crown Hardware contractor network
• → Wire actual CRM features (today the page is the marketing surface; the actual CRM uses ReactCRM patterns from Mac Septic)
• → Add drone-photo estimation flow (Vercel AI SDK + photo segmentation model)
• → Insurance claim auto-sync from FL/TX DOI public adjuster filings
• → Cross-sell to Storm Strike Dispatch lead-buyers as a CRM upgrade

• ! Page is marketing-only today; the actual CRM build will be 4-8 weeks of work once a paying pilot signs
• ! Roofer-vertical SaaS market is competitive — differentiation depends on the bundled data layer, which is real but needs sales training to communicate

The locational-truth layer for the portfolio. For any address: who serves it, with what technology, at what speeds. Real-estate platforms, relocation tools, ISP comparison sites, and the internal Rural Service Lead Engine all consume this same API. Free for existing PermitLookup customers; Stripe-gated for external buyers.

Three FastAPI endpoints live on production (Railway + Cloudflare Tunnel): /v1/broadband/lookup returns every ISP serving an address with technology code + max download/upload speeds + low-latency flag + business/residential code; /v1/septic-score/lookup returns the v2 rural_septic_score for any TX address with component breakdown; /v1/rural-leads/county returns a ranked list of rural-septic candidates by county.

All three are backed by the 924-million-row FCC Broadband Data Collection table (116M unique US broadband-serviceable locations, 2,135 ISPs across all 50 states + DC + PR), partitioned by state for sub-second range scans. Address resolution flows: input address → property_sales fuzzy match → block_geoid → BDC range scan → JSON response. Verified production curls: FM 1696 Huntsville returns 14 ISPs and max 5000 Mbps fiber; 5160 FM 55 Italy returns rural_septic_score 85 with "Strong rural-septic indicator" interpretation.

• > 924,014,277 FCC location-provider-technology records
• > 116M unique US broadband-serviceable locations
• > 2,135 ISPs covered
• > 3 production endpoints (broadband/lookup, septic-score/lookup, rural-leads/county)
• > Sub-3s typical response time on sparse-rural tracts
• > State-partitioned for scale (52 partitions: 50 states + DC + PR)

• → Optimize dense-tract response times (Houston downtown timeouts > 5s)
• → Replicate v2 MV to R730-2 for read-replica load balancing
• → Launch broadband.ecbtx.com marketing landing + Stripe checkout
• → Sign first real-estate API pilot customer
• → Extend /broadband/lookup to all 50 states (currently TX-optimized)

• ! Dense-tract performance (some Houston tracts have 2,400 BDC rows; tunnel-proxy throughput becomes the bottleneck)

Automates zoning compliance checks that normally cost $2–5K per parcel from consultants.

Parcel Screen pulls parcel facts (owner, acreage, zoning, general plan) from Esri FeatureServer endpoints, runs SB-684 eligibility logic and SB-9 lot-split yield calculations, and cross-references a 776M-row CA permits database for prior approvals. Every screen produces an audit log so brokers can defend the result.

Phase 1 shipped 2026-05-13 with the core eligibility engine. Phase 2 will add statewide overlay layers (fire, flood, wetlands) and a Mapbox parcel viewer.

• > 776M-row CA permits database cross-reference
• > Phase 1 shipped 2026-05-13

• → Onboard 2–3 beta development partners
• → Phase 2: statewide overlays + Mapbox viz
• → TX/CO geographic expansion

Service-area intelligence for fleet deployment + route optimization across Central Texas, centered on the Bastrop / Cedar Creek base. Covers the full Mac Septic operating footprint (Travis, Hays, Comal, Caldwell, Bastrop, Williamson) and is licensable to other field-service operators with the same regional shape.

A Leaflet-rendered heat map and density overlay of 159K septic permits across Central Texas, centered on the Bastrop / Cedar Creek base and extending across Travis, Hays, Comal, Caldwell, Bastrop, and Williamson counties within a 120-mile fleet radius. Service-area rings (ideal <15mi, mid 15–30mi, far 30–45mi) and a dump-site inventory layer support route and fleet decisions.

The same engine can be repackaged per-operator as a "service territory intelligence" SKU at $500–1K/mo with competitor-density overlays.

• > 159K septic permits visualized
• > 12 Central Texas jurisdictions across Travis, Hays, Comal, Caldwell, Bastrop, Williamson
• > 120-mile fleet radius from Bastrop / Cedar Creek base

• → Productize as standalone licensing tool
• → Add competitor density overlay

First vertical-market CRM fork; proves the verticalization playbook for the broader strategic-accounts thesis.

Crown Hardware is a 40+ year Columbus IN distributor with 800+ inventory SKUs and $2M+ in annual order volume. The fork adds AI order intake (parsing inbound emails into SKU lines), bill-of-materials assembly, and a production board synced to QuickBooks Enterprise.

The app is demo-ready at crown.ecbtx.com. Contract is stalled on Jenny Kay's bandwidth (new-hire training); strategy is to demo (not re-pitch) once she resurfaces.

• > 800+ inventory SKUs
• > $2M+ annual order volume
• > 40+ year established distributor

• → Demo close (Jenny bandwidth-constrained)
• → Phase 2: Salesforce integration + forecasting

Second vertical proof-point for the verticalization playbook; demonstrates repeatability across distinct trades.

H-Man Electrical runs a contractor-enrichment pipeline driven by TDLR (TX Dept of Licensing & Regulation) license lookups, feeding a live dispatch board for the field crew. Modes include tdlr-load, tier0, tier1, push, and all (`/home/will/permit-api-live/scripts/contractor_enrich.py`).

Deploys are frozen since 2026-04-10 because the GitHub webhook is missing on wburns02/HManElectrical. 8 commits are queued (including the Live Wire redesign). Webhook reconnect is the only remaining gate.

• > TDLR contractor enrichment live
• > Live dispatch board

• → Fix Railway webhook → ship 8 queued commits
• → Activate Live Wire redesign
• → Push lead delivery integration

A separate brand from Mac Septic enables non-cannibalizing paid search; the account is capacity-underutilized.

A 12-route static Astro site (home, about, contact, 4 service pages, 4 service-area pages) for the Honey Wagon Services brand in Rock Hill SC. The local phone number is 803-328-1410 (RingCentral ext 105) — distinct from the Mac Septic Rock Hill line at 803-329-1250.

Google Business Profile creation is pending postcard verification (~2 weeks). Once GBP is live, the site unlocks a separate Google Ads account for Rock Hill SC.

• > 12 routes shipped
• > Launched 2026-05-12

• → GBP postcard verification (~2 weeks)
• → Contact form endpoint
• → Customer review aggregation

Eliminates Google Ads pixel dependency for retargeting; first-party tracking respects modern privacy and survives third-party cookie deprecation.

A custom pixel fires on macseptic.com → POSTs to /api/public/v1/track → a 5-minute APScheduler cron identifies abandoned sessions for known returning permit owners → Brevo dispatches one of three templates (pumping_by_city, emergency, default) → an HMAC-signed vid token + 7-day suppression cap prevents over-mailing.

Alembic 126 added the schema. Sender is dannia@macseptic.com. The same first-party pipeline can be extended to permits.ecbtx.com visitors and to SMS as a second channel.

• > Pixel + 5-min cron live since 2026-05-08
• > HMAC-signed visitor IDs
• > 7-day suppression cap

• → A/B test win-back copy
• → Extend to permits.ecbtx.com visitors
• → SMS abandonment channel

Hot-lead inbox prioritization gives inside-sales a ranked queue instead of a shotgun list; the weekly Opus pass surfaces themes that humans miss.

RingCentral webhooks + Twilio SMS pipe into Deepgram nova-3 for transcription, then Haiku 4.5 runs forced-tool-use triage to score every interaction 0–100. If score ≥ 70, Sonnet 4.6 drafts a contextual reply. DNC enforcement runs at the suppression layer so any reply path respects opt-outs.

A weekly Opus 4.6 strategist email summarizes sentiment, surfaces top themes, and recommends campaign adjustments. The first Sunday Opus run was 2026-05-03 06:00 CT against the Dannia campaign `email-openers-spring-2026`. AI spend is hard-capped at $25/day with auto-pause.

• > 90-day backfill verified
• > $25/day AI spend cap with auto-pause
• > Multi-channel: calls + SMS + email + chat

• → Multi-vertical productization
• → Customer-facing scoring dashboard
• → A/B test reply tone variants

Google Ads is the primary inbound channel; SEO baseline is essentially zero (0/38 ranking). A 100-day window gives 30–40 days of ranking visibility before the next investor checkpoint.

Weekly `python3 -m scripts.ai_rank_check` measures position on 38 commercial keywords (e.g., "septic pumping Nashville", "emergency septic repair TN"). Execution levers are on-page optimization, a content calendar, link velocity from honeywagonservices, and AI copywriting for keyword-targeted landing pages.

The goal is top-10 on 10–15 commercial keywords in the 100-day window (2026-05-11 → 2026-08-19). A mid-window pivot is planned if no movement appears by day 60.

• > 38 commercial keywords tracked
• > 100-day execution window 2026-05-11 → 2026-08-19
• > Weekly ranking checks via gpt-4o

• → Ship 5 keyword-targeted landing pages
• → Backlink velocity from honeywagonservices
• → Mid-window pivot if no movement by day 60

Demonstrates life-sciences literacy + a founder personal story; resonates with investors interested in founder DNA + healthspan.

MyHeritage 720K SNP sequencing identified 16 identical alleles shared between Will and his father, including ACTN3 (endurance), COMT Val/Val (warrior brain), and BDNF (exercise response). The site is structured as three editorial sections: "Fifty Years of Training", "Warrior Brain", and "Built to Live Long".

A PSMA-PET scan was performed May 14 (results in ~2 weeks). DNA from siblings (Nikki, Chris, Matt) is pending. Color Health expansion is queued for actionable variants.

• > 720K SNP MyHeritage sequencing
• > 16 identical father–son genotypes identified
• > Multi-section editorial narrative

• → Integrate sibling DNA when available
• → Color Health expansion for actionable variants
• → Re-publish post-PSMA-PET results

Code violations are the strongest leading indicator of forced repair, sale, or service spend. Joined to the permit graph this becomes a national distress-signal dataset usable for collections, real estate, and field-service lead gen.

A 297-million-row code-violations table loaded into the same permits database on the T430 node. Each record carries jurisdiction, parcel, violation type, status, and date. When joined to property_sales (97M), property_liens (37M), and the permit graph (777M), it forms a national distress-signal layer.

Loaded incrementally from county and municipal portals via the gov-data-extraction framework. The dataset is the basis for a forthcoming "distress leads" product targeting field-service contractors and real-estate distress investors.

• > 297,152,967 rows
• > 101 GB on-disk
• > Joined to property_sales, property_liens, permits

• → Build distress-signal API endpoint
• → Pilot with one Texas roofer or one collections firm
• → Add weekly refresh cron from top 25 county portals

Property records turn permits into decisions. Owner names, sale prices, lien histories, and assessed valuations make every permit row actionable as a sales lead or distress signal.

Four tables form the property side of the platform: property_sales (97M rows, 23 GB), property_liens (37M rows), acris_parties (33M NYC ACRIS records), and property_valuations (9M rows). All four are the join targets for the enrich_norm_addr.sql and enrich_with_sales.sql cron jobs that hydrate hot_leads with owner and price data each morning.

Original acquisition was a mix of BatchData purchases, county assessor scrapes, and the NYC ACRIS public bulk feed. The combined suite means any permit address in the platform can be enriched with a real owner name, last sale price, lien status, and assessed value in milliseconds.

• > 97M property sales records
• > 37M property lien records
• > 33M NYC ACRIS party records
• > 9M property valuation records
• > Joined nightly to permits + hot_leads via enrichment SQL

• → Add owner-phone enrichment via contractor_licenses join
• → NYC mortgage-doc subset for distressed-property cohort
• → Public API tier with usage caps

Removes per-token AI spend from internal triage, scoring, and summarization workloads. Owns the inference layer rather than renting it. Frees the AI Interaction Analyzer and OCR Pipeline to run at any volume without a cost cap.

A locally-hosted LLM cluster running on the R730 (88 cores / 768 GB RAM, dual NVIDIA GPUs, GPU-aware fan control). The active reasoning model is Qwen 3.5 122B at 81 GB on disk; nomic-embed-text handles embedding workloads for the RAG service in security-stack. Both are served by Ollama and reachable across the Tailscale mesh.

The cluster is the substrate for any non-real-time AI workload: nightly summary jobs, batch transcription scoring, embedding generation for the SOC RAG, OCR post-processing on tank inspections, and reasoning over scraped permit corpora. Anthropic Claude is reserved for latency-sensitive customer-facing flows; everything internal flows here.

• > Qwen 3.5 122B (81 GB) live since 2026 Q2
• > Dual-GPU R730 with custom fan control
• > nomic-embed-text for vector workloads
• > Tailscale-mesh accessible from every node in the cluster

• → Add a third smaller model for low-latency triage
• → Move PeakDipVibe pipeline fully on-prem
• → Publish internal usage dashboard

Most counties refuse PIA requests for OSSF data or quote thousands of dollars. Tank Vision turns the freely available PDF permits into structured rows for the same outcome at zero marginal cost per county.

Tank Vision is a self-improving OCR pipeline at /home/will/ocr-pipeline. It takes county septic-permit PDFs (currently Comal County CCEO in Texas), runs dual-GPU OCR, applies a learned-rules extractor that grows over time (learned_rules.json, learned_examples_comal.json), and writes structured rows directly into the permits database.

The pipeline is the cost-avoidance answer to county PIA requests like the Kendall County $3,456 quote (declined April 2026). Each new county adds incremental scrape effort but zero ongoing per-record cost.

Active coverage now spans Texas + Tennessee: Comal County (TX) plus a Tennessee cohort across Hamilton (Chattanooga), Davidson (Nashville), Rutherford (Murfreesboro), Shelby (Memphis), Williamson (Franklin), Wilson (Lebanon), Maury, and a Johnson City 4-county bundle. Hays, Williamson (TX), and Bexar are next in queue.

• > Comal CCEO septic permits actively processed (TX)
• > TN coverage: Hamilton (Chattanooga), Davidson (Nashville), Rutherford (Murfreesboro), Shelby (Memphis), Williamson (Franklin), Wilson (Lebanon), Maury, plus Johnson City 4-county bundle
• > Dual-GPU launch script (launch_dual_gpu.sh)
• > Self-improving learned rules + examples corpus

• → Add Hays + Williamson (TX) + Bexar counties
• → Generalize learned-rules engine across counties
• → Auto-detect and queue new permits from county portals

Turns raw permit records into qualified sales leads automatically. Joins property sales, exact-and-fuzzy-matches contractors and applicants, scores leads, and pushes only verified enrichments to the primary node. The cost-tracked, anomaly-gated design means the LLM spend is auto-bounded and silent regressions are caught.

A standalone orchestrator at /home/will/permit-enrichment-agent on R730-2 that runs three tiers of enrichment over the permit graph.

Tier 1 is pure SQL: address normalization, exact-match joins against property_sales and contractor_licenses, PostgreSQL trigram fuzzy matching with auto-accept above 0.7 and an escalation queue between 0.4 and 0.7. A typical run enriches ~4,600 leads in under 10 minutes.

Tier 2 escalates the trigram-ambiguous matches to Claude Haiku in batches of 20; Haiku accepts above a tuned confidence of 0.85 and rejects below 0.4. Anything in between escalates again to Claude Sonnet with a 0.7 accept threshold. Every LLM call is logged with token counts and dollar cost in a SQLite ledger (costs.db) for per-day and per-month budget reporting.

Tier 3 is rich Sonnet analysis on the highest-value leads, plus a description miner (Haiku-only) that extracts septic-tank size and inspection-status hints from free-text permit descriptions. The agent also runs an anomaly detector that compares today's tier-1 stats to the 7-day median and fires a RingCentral SMS to the operator on underflow (today=0 vs median 1 already caught a regression on 2026-04-24).

The nightly sync pushes enriched rows from R730-2 to T430 with a regression gate: if any field has lower coverage on R730-2 than T430, the sync aborts and alerts. Twilio is used for phone validation on top-priority phones; a website finder enumerates contractor URLs and classifies them. The whole thing is flock-guarded so cron drift can't fire two instances.

Cron is paused since 2026-04-24 after a Tier-2 batch returned 2,250 consecutive Anthropic API errors. The Tier-1 SQL pipeline is independently working and ready to re-enable; Tier-2 needs API stability work or a fallback to the on-prem Sovereign LLM cluster (Qwen 3.5 122B) to bypass external API risk.

• > 4,653 property-sales joins per run on 2026-04-24 (last live run)
• > ~2,000 contractor + 250 applicant trigram escalations per run
• > 3-tier LLM pipeline: trigram → Haiku → Sonnet with tuned thresholds
• > SQLite-backed cost ledger with per-day and per-month budget tracking
• > RingCentral SMS alerts to +1 979-236-1958 on anomaly underflow
• > Cross-node sync with regression gate (R730-2 → T430)
• > flock-protected single-instance guarantee

• → Re-enable Tier-1 SQL cron (works standalone, no external API)
• → Migrate Tier-2 fuzzy matching to local Qwen 3.5 122B (Sovereign LLM Cluster) to eliminate external API risk
• → Productize as a per-vertical enrichment SKU once stable

• ! Tier-2 Anthropic API instability caused 2,250-error cascade on 2026-04-24 → cron killed
• ! No git remote — code lives only on R730-2; back up or move to wburns02/permit-enrichment-agent

Owns the entire detection-and-response stack instead of renting it. Provides centralized monitoring across every node in the portfolio, plus a pentest workbench and AI-assisted alert triage that would otherwise require an MSSP.

A full self-hosted security stack at /home/will/security-stack with three composed pieces: Wazuh SIEM (1514/1515), Elasticsearch (9200), and a RAG service backed by Qdrant for AI-assisted alert reasoning. A SOC dashboard fronts the alert queue and lets the operator query the RAG with natural language ("which hosts ran sudo today and have an unfamiliar process tree").

A dedicated pentest folder holds offensive tooling for periodic internal exercises. The whole stack runs in Docker Compose with separate compose files for SOC services and RAG service.

• > Wazuh agents on every Tailscale node
• > Elasticsearch index for log search
• > Qdrant vector DB for RAG-assisted triage
• > SOC dashboard + pentest tooling

• → Onboard one external pilot client
• → Productize as a "managed detection light" SKU
• → Auto-generate weekly threat-summary email

Live B2C SaaS with HTTP-only JWT auth, encrypted journals, partner pairing, and AI-generated relationship insights. Demonstrates productized consumer pattern and is launch-ready for subscription monetization.

DrPhil is a couples relationship wellness app at drphil.ecbtx.com. Each partner pairs by invite code and runs through a relationship questionnaire onboarding, then into daily mood + relationship check-ins. Mood and journal entries are encrypted at rest with Fernet; auth runs on HTTP-only JWT cookies. The dashboard shows mood trends with AI-generated insights from Claude, 8 guided couple exercises with completion tracking, daily relationship questions, weekly intentions, a partner activity feed, 33 curated date-night ideas with filters, and nudge reminders. The app is a PWA with a service worker so it installs on mobile home screens. Built on FastAPI async + PostgreSQL on the backend, React 19 + Vite + Tailwind 4 on the frontend, deployed via Railway nixpacks auto-deploy from main.

• > Live on drphil.ecbtx.com
• > Fernet-encrypted journals
• > JWT cookie auth
• > PWA installable
• > 8 guided couple exercises
• > 33 date-night ideas
• > Partner pairing by invite code

• → Stripe paywall ($5-10/mo)
• → Mobile push notifications
• → Premium tier with audio-guided exercises
• → Couples therapist referral marketplace

Live public dashboard at stocks.ecbtx.com with a fully automated daily data pipeline. The peak/dip scanner finds post-earnings gap-down re-entry signals; AI enrichment runs on the Sovereign LLM Cluster (zero per-token cost).

PeakDipVibe is a market analytics dashboard at stocks.ecbtx.com powered by a daily ETL pipeline. A systemd timer runs Mon-Fri 15:30 CT to refresh OHLCV data via yfinance + Stooq + FMP into a 148 MB SQLite market.db. The peak/dip scanner identifies earnings gap-downs where fundamentals justify a re-entry, with a tunable pounce score for trade readiness. Earnings catalysts are auto-generated by the local Qwen 3.5 122B model running on the Sovereign LLM Cluster (zero per-token cost), with weekly recalibration on Fridays and monthly What-I-Learned AI reviews on the 1st. Coverage: S&P 500 + NASDAQ-100 + Dow 30 + the broader NASDAQ. Frontend is React 19 + Vite + Tailwind 4 + Recharts on Railway; backend FastAPI is on R730 via Tailscale.

• > 500+ tickers covered
• > 78 earnings events / week tracked
• > 50M+ OHLCV rows in market.db (148 MB)
• > 22 automated daily refreshes / 30d
• > On-prem Qwen 3.5 122B for case-study generation
• > Mon-Fri 15:30 CT systemd timer

• → Paywall premium signals / earlier alerts
• → Add weekly newsletter from monthly AI review
• → Mobile push for peak/dip events
• → Trade simulator monetization

Launching real-world use shortly. The leads view filters dumpster prospects by pool / E4 / permit signal off the 2.5B-row data cluster; the gate captures sessions before lead surface. Live funnel analytics (visitors / sessions / funnel stages / hot leads / live feed) refresh every 5 minutes via a Tailscale-Funnel JSON drop, giving operators real-time visibility into pre-launch warm-up.

Dumpster Leads is a roll-off dumpster lead-gen product at dumpster.ecbtx.com, hosted in the permitlookup-web Next.js 16 / React 19 / shadcn app on Vercel. The customer-facing leads dashboard filters dumpster prospects by pool, E4, and permit signal off the 2.5B-row cluster, with mobile-first responsive design and a password gate to wall off pre-launch beta access. A per-session event tracker writes to a dumpster_events Postgres table on Tailscale (event types: gate_unlock, page_view, filter_apply, lead_click, phone_tap, plus smoke-test variants) with full visitor + session + geo + viewport instrumentation. An admin analytics dashboard at /admin/dumpster-analytics renders KPIs (visitors today, sessions today, events 24h, last event), a 30-day series chart, geo distribution, the 5-stage funnel, hot-leads table, and a live event feed. The full rollup is refreshed every 5 minutes by a server-side cron (dumpster_analytics_dump.sh on R730) that runs the SQL and writes a single JSON snapshot to dashboard-cache served verbatim via Tailscale Funnel; Vercel proxies that JSON via /api/admin/dumpster-analytics. The architecture is the canonical pattern: customer-facing Next.js on Vercel, data on Tailscale-internal Postgres, analytics by static-JSON proxy — fast, secure, zero DB exposure.

• > Live on dumpster.ecbtx.com (Vercel)
• > Backed by 2.5B-row permit + property cluster
• > 5-minute analytics refresh cron
• > Full funnel instrumentation (gate · view · filter · click · phone)
• > shadcn / Tailwind 4 design system
• > Tailscale Funnel JSON proxy for analytics

• → Sign first roll-off pilot customer
• → Stripe per-lead pricing
• → Outbound script for first 10 dumpster operators
• → Geo expansion (FL, GA, NC, SC)

• ! Pre-launch — funnel needs first real-world traffic to validate conversion math

Personal-health analytics with pattern recognition, Real Age scoring, training/sleep/weight tracking, and an AI health agent. Active development. Foundation for the Walter DNA Project healthspan thesis productized.

HealthForge is a personal-health platform at health-api.ecbtx.com. It pulls biometric streams from Whoop, Garmin, and Apple Health, then surfaces a Morning Briefing (sleep quality + stress + weight + RHR), a 7-day Lowest HR tracker with why-logging, a Real Age curve, weekly training schedules with daily set logging, and an AI Health Agent (Claude) with persistent conversation. Insights are computed from 30+ pattern-recognition rules that surface lowest-HR days, training-stress anomalies, sleep-quality dips, and manual-weight gaps. Backend is FastAPI + PostgreSQL on Railway; frontend is React 19 + Vite + Tailwind 4 with Recharts visualizations. The system is the operator personal health log and the technical seed for any future healthspan SaaS productized off the Walter Burns DNA Project.

• > 18 commits / 30d
• > 7 pages
• > Wearable integrations: Whoop + Garmin + Apple Health
• > AI Health Agent (Claude) with conversation persistence
• > Real Age trend computation
• > 30+ insight pattern-recognition rules
• > Weekly training schedules with set logging

• → Productize alongside Walter DNA Project
• → Pricing tier launch
• → Community / leaderboard layer
• → Clinical labs integration

Healthcare consulting platform with OCR ingestion of clinical PDFs, project + contractor + client management, compliance audit logging, quality metrics dashboards, and training module tracking. Targets healthcare consulting firms and medical-records vendors.

KDJ Clarity (formerly opus46) is a medical-records quality-assurance platform for Texas-based clinical documentation consulting. The frontend ships 15 pages: Dashboard, Projects, ProjectDetail, RecordQueue, Quality, Contractors, ContractorDetail, Clients, Compliance, Billing, Reports, AuditLog, Training, Settings, Ingestion. The flagship workflow is PDF OCR ingestion — automated PDF-to-record extraction with a nurse review queue and compliance audit logging on every change. Stack: React 19 + Vite 8 + Tailwind 4 + Framer Motion + React Router 7 + Lucide React. Seed data carries 12 clients, 8 projects, and 42 contractors. Built for KDJ Consultants internal use; productizable to other healthcare consulting firms.

• > 15 pages
• > 12 seed clients
• > 8 seed projects
• > 42 seed contractors
• > PDF OCR ingestion pipeline
• > Compliance audit logging
• > Training module tracking

• → Sign 2-3 KDJ-adjacent consulting clients
• → Multi-tenant data isolation
• → HIPAA BAA path

• ! HIPAA claims explicitly removed from earlier versions; needs real compliance review before sales

Hospital-grade staffing platform with HIPAA-compliant MFA, RBAC, audit logging, idle timeout, real-time acuity-driven recommendations, and a Claude-Haiku clinical AI assistant. 50 ops pages across shifts, acuity, hiring, retention, payroll, RRT, productivity. Productizable to mid-large hospital systems.

NurseStation is a hospital nurse-station operations platform deployed on Railway. The frontend ships 50 pages covering shifts, staff, analytics, acuity, census, briefing/handoff, payroll, hiring, scheduling, float pool, agency staffing, retention intelligence, training, onboarding, overtime, rapid response, productivity, experience, skills, LiveOps, incident hub, recognition, SafeHours, and 20+ more. Acuity data is modeled across 7 units (ICU, CCU, MS-A, MS-B, Oncology, Telemetry, ED) with 40-200 patient records per dataset and 2.5-9 nursing care hours by acuity level. The platform is HIPAA-hardened: MFA, role-based access control, audit logging for every PHI access, idle timeout, session expiry, and CSP headers. A Claude Haiku AI clinical assistant streams via an Express 5 proxy. Built with React 19 + Vite 8 + Tailwind 4 + Framer Motion.

• > 50 operational pages
• > 7 modeled units
• > 64 commits / 90d
• > HIPAA MFA + RBAC + audit log + idle timeout
• > Claude Haiku streaming AI assistant
• > Production Railway deployment

• → HIPAA BAA + SOC 2 path
• → Pilot 1-2 mid-size hospitals
• → EHR integration (Epic / Cerner)
• → Multi-tenant data partitioning

• ! No HIPAA BAA yet — sales blocker until executed
• ! Healthcare procurement cycles 6-18 months

Full-featured commercial-real-estate / enterprise-sales CRM with kanban-by-deal-stage, 7-tab engagement view, document OCR, and AI deal alerts. Built navy + copper brand. Ready to pilot.

EminenceCRM is a commercial-real-estate / enterprise-sales CRM. The product covers the full deal lifecycle: kanban pipeline by deal type and stage, a 7-tab engagement detail page (overview, activities, tasks, documents, notes, history, timeline), document OCR for inbound deal materials, click-to-edit tasks and activities, real-time notifications, and AI-powered deal alerts. The data model carries 50+ prospects, 20+ companies, 30+ contacts, 100+ tasks, 50+ activities, and 40+ documents in seed data. Frontend is React 19 + Vite 8 + Tailwind 4 + TanStack Query + React Hook Form + Zod + dnd-kit + Recharts + Zustand + Playwright. The whole stack ships behind a Caddy reverse proxy on Railway with VITE_API_URL backend wiring.

• > 13 feature modules
• > 7-tab engagement view
• > dnd-kit drag-drop kanban
• > 21 commits / 90d
• > Document OCR pipeline
• > Production Railway + Caddy deployment
• > Real-time activity sync

• → Sign 1-2 commercial-real-estate pilot teams
• → AI deal-summary generation
• → Outlook / Gmail email sync
• → QuickBooks revenue tie-out

Order-tracking pipeline for CNC Federal Supplies, a GSA / federal-supply dropship reseller (Clayton Trammell, clayton@cncfederal.com). Pulls Red Falcon CSV exports from the GSA portal, syncs QuickBooks Online Sales CSV, and uses Claude Sonnet 4 Vision to OCR vendor invoices, customer POs, and shipping documents into a single order timeline. Solves the dropship-reseller paperwork bottleneck end-to-end.

CNC Federal Order Tracking is a focused workflow tool for federal-supply dropship resellers. The customer is CNC Federal Supplies (Clayton Trammell, clayton@cncfederal.com). The data model is order-centric: users / orders / order_items / order_notes / order_documents / import_logs — not machines, work orders, or operators. Three ingestion paths feed the pipeline: Red Falcon CSV exports from the GSA portal (order intake), QuickBooks Online Sales CSV (revenue reconciliation), and Claude Sonnet 4 Vision PDF OCR for vendor invoices, customer POs, and shipping/tracking documents. The frontend ships 6 modules: auth, dashboard (kanban-style order pipeline + KPI cards + searchable orders), import (CSV preview + PDF OCR upload), orders, reports, and settings. Built with React 19 + Vite 8 + Tailwind 4 + dnd-kit + Recharts + date-fns + Playwright E2E coverage. Multi-stage Docker build (Node 22) for production deployment; VITE_API_URL env wires the backend.

• > Order-centric schema: users / orders / order_items / order_notes / order_documents / import_logs
• > Red Falcon CSV (GSA portal) intake
• > QuickBooks Online Sales CSV sync
• > Claude Sonnet 4 Vision OCR on vendor invoices + customer POs + shipping docs
• > Multi-stage Docker production build
• > Playwright E2E coverage

• → Backend API for live order data (currently mocked)
• → Generalize beyond CNC Federal to other GSA dropship resellers
• → Shipping-carrier tracking webhook ingestion (FedEx / UPS)
• → QuickBooks Online deeper sync (invoices, not just sales CSV)

• ! Backend API not wired yet — orders are mock data until backend ships
• ! Single-customer product today; productization depends on Clayton validating the workflow

End-to-end craft-brewery platform: 14 sequential feature pipelines covering Ferment Lab (real-time fermentation monitoring), Brew Cost Lab (recipe → cost), POS loyalty + live tap menu, wholesale orders, beer ratings, event ticketing, Birthday Hub, batch cost calculator. Frontend + backend both production-shaped.

BreweryCRM is a craft-brewery operations platform with 14 sequential feature pipelines. The frontend covers a real-time Ferment Lab (sensor-driven fermentation monitoring), Brew Cost Lab (grain-to-glass recipe cost analysis), POS loyalty points auto-award, live tap menu board with dynamic pricing, wholesale order workflow with batch invoicing, beer ratings/reviews, event ticketing, Birthday Hub for customer loyalty check-ins, batch cost calculator across brewing phases, and a schedule summary calendar. The backend (Bearded Hop API) ships 44 SQLAlchemy models and 32 FastAPI route modules covering recipes, vessels, brew days, quality checks, customer notes, visit history, keg inventory, and production dashboards. Built with React 19 + Vite 7 + Tailwind 4 + TanStack Query + Recharts + Playwright (frontend) and FastAPI 0.115 + SQLAlchemy 2.0 async + asyncpg + Alembic + Pydantic (backend). Deployed on Railway.

• > 14 feature pipelines
• > 44 SQLAlchemy models
• > 32 API route modules
• > Detailed seed data: recipes, kegs, vessels, customer notes
• > Playwright E2E test coverage
• > Railway deployment ready

• → Sign first paying brewery
• → POS hardware integration (Toast, Square)
• → Distributor / wholesale portal
• → TTB (federal alcohol) reporting

• ! Dormant since March 2026 — needs launch effort or single-pilot revival

Full-featured landscape-service platform: 20 major modules covering smart routing, BI, weather-integrated scheduling, AI revenue upsell, property-centric service memory, dispatch map, invoice dunning, safety/compliance, online booking, integrations hub, dynamic pricing, vendor PO. Plus a pre-loaded 272K Austin lead database in the API for direct-mail targeting.

LandscapeCRM is a 20-feature landscape-service operations platform. Notable modules include Smart Route Planner (drag-drop crew scheduling), Business Intelligence (revenue/profit/forecasting), Weather Command (weather-integrated smart scheduling), Smart Estimator, Crew Field Hub (mobile-first ops), Revenue Autopilot (AI upsell discovery), Customer Engagement Hub (comms timeline + health scoring), Property Intelligence (property-centric service memory with chemical compliance tracking), Daily Closeout (5-minute end-of-day review), Live Dispatch Map, Invoice Follow-Up Center (dunning sequences), Safety & Compliance (certifications + incident logging), Service Request Manager (online booking), Integrations Hub (QuickBooks/Stripe/Google Calendar + 5 more), Service Price Book with tiers/margins, and Vendors & Purchase Orders. The backend carries a 272K-row Austin lead database for direct mail prospecting, 19 SQLAlchemy models, 23 FastAPI route modules, plus direct mail campaign models and a bulk prospects import endpoint. Stack: React 19 + Vite 7 + Tailwind 4 + TanStack Query + React Router 7 + Recharts (frontend); FastAPI + SQLAlchemy async + asyncpg + Alembic (backend); Railway deployment.

• > 20 feature modules
• > 272K Austin lead database
• > 19 backend models
• > 23 backend route modules
• > Weather integration
• > AI revenue upsell engine

• → Sign first paying landscape company
• → Use 272K Austin DB for outbound + first-customer acquisition
• → QuickBooks tie-out
• → Mobile crew app

• ! Dormant since March 2026 — needs launch effort

Experimental gameplay-engine R&D: deterministic Mulberry32 PRNG, full pitch/contact/fielding/baserunning physics, manager AI + lineup builder, 30-season franchise mode with dynasty + Hall of Fame, minor leagues, draft + trades + free agency. Out-of-the-Park Baseball-style niche with a tighter game loop.

Claude Ball is a browser-based baseball franchise simulator built as an experimental gameplay engine. The core engine handles pitch, contact, fielding, baserunning, and at-bat resolution with deterministic Mulberry32 PRNG (every game is replayable from a seed). Manager AI builds lineups and makes in-game decisions. Franchise mode supports 30+ seasons with dynasty tracking, Hall of Fame induction, minor leagues, draft, trades, free agency, playoff brackets, box scores, play-by-play, and league standings. Player + team editors enable custom leagues, and a career mode includes childhood-stage selection (Living Dynasty) with personality point budget. Stack: React 19 + TypeScript 5.9 + Vite + Tailwind 4 + Pixi.js 8 (2D rendering for baseball diamond) + Zustand + Recharts + better-sqlite3 + Express + Playwright. Local dev at http://localhost:5173 (Vite) and http://localhost:3000 (Express); not yet public.

• > Deterministic Mulberry32 PRNG
• > Full at-bat / fielding / baserunning physics
• > 30-season franchise mode
• > Minor leagues + draft + trades + free agency
• > Pixi.js 8 2D engine
• > Career mode with Living Dynasty mechanic

• → Sprite + gameplay-only sprints (Claude Ball GFX branch)
• → Public web playable demo
• → Decide commercial vs portfolio-only positioning

• ! Not deployed publicly
• ! No commercial path defined yet

Force multiplier behind the entire portfolio. Spawns parallel Claude Code workers, detects completion via promise tags, runs project memory logging, and monitors production sites with auto-fix triggers. The reason one operator ships across 32 products.

Conductor AI is the internal orchestration layer for the portfolio. It spawns autonomous Claude Code worker processes in parallel, detects completion signals via promise-tag parsing on streamed JSON output, manages task queues, and runs a Rich TUI dashboard plus REST API on port 8787 with an embedded PTY terminal. Conductor monitors production sites (drphil.ecbtx.com, react.ecbtx.com, others) and triggers auto-fix sprints when health checks fail. The Gen4 experiments add a blackboard for cross-agent state sharing, a metacognition layer, adversarial verification of outputs, and a context distiller for long-running sessions. Conductor is integrated into Claude Terminal for one-command project switching. Stack: Python 3.14 + FastAPI + Uvicorn + SQLite (async) + Rich + Typer + Claude CLI subprocess workers.

• > N-parallel Claude Code worker pool
• > Promise-tag completion detection
• > Production-site health monitoring with auto-fix
• > Gen4: blackboard + metacognition + adversarial verifier + context distiller
• > Embedded PTY + web dashboard on port 8787
• > Project memory logging

• → Open-source the orchestrator
• → Publish Gen4 experiment papers
• → Multi-operator deployment

The Conductor AI dashboard frontend (GitHub: wburns02/conductor-dashboard). Layout, navigation, and branding are hardcoded to Conductor. Hosts a live xterm.js terminal emulator with Framer Motion animations and a session launcher (TerminalPage) for spawning Claude sessions across the project surface. Not a shared library — this is the operator UI for Conductor AI itself.

This repo is the Conductor AI dashboard frontend. GitHub remote is wburns02/conductor-dashboard. The package.json is private: true, version 0.0.0, with no exports field; per the deep-doc audit there are zero downstream import consumers. Layout.tsx hardcodes Conductor branding and navigation; the "project launcher" in src/pages/TerminalPage.tsx is a menu for spawning Claude sessions across the Conductor-managed surface (DrPhil, NurseStation, ClaudeBall, PermitAPI, Conductor AI, ClaudeBallGFX), not a list of products that import this library. Highlights: live xterm.js terminal emulator with Framer Motion animations and web-links addon, advanced animation patterns (performance sprint, circuit-breaker UI), and a component framework under src/components and src/features. Stack: React 19 + TypeScript + Vite 7 + Tailwind 4 + Framer Motion 12.34 + xterm.js + React Router DOM 7. Docker-containerized via Node serve on port 3000; Playwright test coverage.

• > Conductor AI operator UI (single product, not multi-tenant)
• > xterm.js terminal emulator
• > Framer Motion 12.34 animation primitives
• > Session launcher (TerminalPage) for spawning Claude sessions
• > Performance sprint patterns + circuit-breaker UI for service health

• → Tighten coupling with the Conductor AI backend session API
• → Surface real-time agent status + cost telemetry in the dashboard
• → Per-session transcript replay and export

Internal SEO automation engine processing content for portfolio properties at scale. Multi-container stateless design with nginx reverse proxy. Drives the SEO Rank Push initiative and the 19-page county-specific landing pages without per-page agency cost.

ECBTX SEO Service is an internal Docker-containerized microservice processing SEO content at scale for portfolio properties. The architecture is stateless with separate data, models, and credentials directories; nginx serves as reverse proxy / load balancer for multi-tenant or multi-endpoint routing. The service powers content automation for permit lookup, county-specific landing pages, and other SEO-heavy properties in the portfolio. Deployed on R730 with Docker Compose; version managed via version.txt; credentials isolated.

• > Multi-container Docker stack
• > nginx reverse proxy
• > Stateless processing layer
• > Powers 19+ county landing pages + main brand sites

• → Externalize as managed SEO content engine SKU
• → Publish content velocity dashboard
• → Add programmatic SEO scaling per vertical

Local-only, single-operator design pipeline that turns parameter sliders into print-ready STLs that have already been bed-fit checked, slicer-validated, and visually previewed before they hit the printer.

STL Generator is a parameter-driven 3D-print pipeline at stl.ecbtx.com (HTTP Basic Auth gated, local-only). Five generator modules cover distinct print categories: Clicker for fidget mechanisms with tunable click force and travel, Topo Map for 3D-printed terrain built from DEM tiles via external elevation APIs and Nominatim place lookup, Flexi Zoo for hinged-segment articulated animals, Dragon Builder for procedural dragons with limb and scale parameters, and History for re-running and forking prior parameter sets.

Every generated mesh runs through a four-gate validation pipeline before it reaches the slicer. V1 is a trimesh watertight + manifold check. V2 enforces bounding-box limits against the Qidi X-Max II build volume (300x250x300mm) and bed-fit. V3 invokes PrusaSlicer 2.9 headless to produce real gcode and confirms the slicer accepts the geometry. V4 renders four-angle PNG previews via pyrender on EGL and computes perceptual hashes so visually broken meshes are caught before plating.

Storage policy keeps STLs, gcode, and render PNGs on /mnt/win11/Fedora rather than the home drive, per the operator drive map. Compute is fully local with zero per-print cost. The target printer is the Qidi X-Max II.

• > 5 generator modules (Clicker, Topo Map, Flexi Zoo, Dragon Builder, History)
• > V1 trimesh watertight + manifold validation
• > V2 bed-fit + bbox check vs 300x250x300mm build volume
• > V3 PrusaSlicer 2.9 headless gcode validation
• > V4 pyrender 4-angle PNG previews + perceptual hashes
• > External DEM elevation + Nominatim place lookup for Topo Map

• → Add more procedural generators (terrain variants, mechanical kits)
• → Slicer profile auto-detect per material
• → Multi-printer support beyond Qidi X-Max II
• → Public hobby SaaS tier if external interest emerges