Pick the artifact, get it shipped.
- AI POC
- UX / service‑design consult
- Architecture gut‑check
- Audit
- Feasibility study
- Weird R&D
Recently shipped
AssetTrax / Dataref — VP of AI. Brought computer-vision + industrial RAG from prototype to production. Lane: AI POC.
Founders with a stuck POC. Heads of AI, product, or data. Execs deciding whether to fund the bet.
Three reasons buyers land here. One: a POC stuck between demo and production — the prototype's brittle, engineering can't get past it, the board's asking when revenue arrives. Two: an AI bet you need a builder to validate before you commit the budget. Three: a "what if" sitting in a strategy deck that deserves a working prototype to prove or kill. I take any of those from where they are to a real, defensible system in production — or tell you straight if it shouldn't ship.
Most of this is agentic AI right now, but the lane is broader: devices, apps, edtech, marketing-as-public-art, anything early-web in spirit. Cut my teeth in advertising and marketing during the early web; I've been showing up to brand-new categories ever since.
- Industrial-grade RAG systems — Neo4j knowledge graph + pgvector / Pinecone, hybrid retrieval, semantic re-ranking
- Multi-agent orchestrators on GCP Cloud Run + Pub/Sub (or AWS Lambda + EventBridge / SQS)
- Automated research & lead-intel platforms in N8N — 49 workflows, self-evolving query engine, in production today
- Scalable computer-vision pipelines with regression test harnesses — YOLO, OpenCV, custom CNNs, edge deployment
- Document ingestion & classification at scale — Gemini, Claude, OpenAI; OCR + EXIF + WebP pipelines
- Cost-engineered LLM caching layers — 60%+ reduction observed in production
- USPAP-compliant valuation pipelines with hash-chained, court-ready evidence trails
- OpenTelemetry distributed tracing · circuit breakers · dead-letter queues · Terraform IaC
- Non-AI R&D — devices, apps, UX prototypes, ed-tech reinvention, marketing-as-public-art disruption work, anything early-web in spirit. POCs aren't an AI thing for me — they're a temperament.
- Advisory — scaling POCs to production, viability-of-concept reviews, team & hiring plans, vendor / model selection, eval-framework design, AI roadmaps for execs without a CTO yet. Right-sized for a single engagement or a standing monthly retainer.
Live in production right now: a 49-workflow autonomous agent platform with self-evolving query logic, hybrid retrieval over a knowledge graph, and multi-model inference. Hitting 95%+ classification accuracy at 100K docs/day. 60%+ LLM cost reduction. 50+ appraisers on the Appraisal Dream waitlist.
Stack details
Software 3.0, end to end. Ralph-Wiggum-style autonomous loops. Multi-agent orchestration via OpenClaw runtime. Tool use + Claude Skills. Custom eval & regression harness. Multi-model inference across Claude, Gemini, OpenAI, and open-weight (Hermes / Llama). Industrial-grade RAG over a Neo4j knowledge graph + pgvector / Pinecone hybrid retrieval. MCP-aware agent design.
For heads of AI, product, data, or engineering at insurers, banks, valuation firms, RIAs, and F500 business units. For founders with a POC that needs production hardening. For execs deciding whether to fund the AI bet — get a builder's read first.