Houston AI Developer — Custom Models for Texas Businesses

Houston AI demand — what I see in the metro

The Houston AI buyer pool is unusual. The Texas Medical Center generates more healthcare AI demand than any city outside Boston, but most of that demand is gated by HIPAA and risk-management constraints that rule out third-party LLM APIs. The Energy Corridor and Houston Ship Channel generate operations-AI demand at scale — predictive maintenance, claim review, contractor-document extraction — but with data classification rules that require self-hosted models. Microsoft-shop enterprises across the metro want AI features inside the .NET / Azure AD perimeter they already operate, not bolted onto an OpenAI account that lives outside their compliance posture. All three converge on the same shape: custom AI built by someone who will run it inside the buyer\'s perimeter.

I work as that engineer. The model is fine-tuned on the buyer\'s corpus. The inference runs on the buyer\'s hardware (or the buyer\'s Azure tenant). The eval harness, monitoring, and deploy runbook all belong to the buyer. There is no third-party API in the production path. There is no per-token fee compounding for the lifetime of the application.

The trade is up-front capex for zero-marginal-cost inference at scale. For Houston buyers whose monthly OpenAI / Anthropic / Bedrock bill has crossed $20K, the economics tilt before the engagement ends. For buyers running compliance-gated workloads, the trade is non-economic — the third-party path is simply not allowed.

What I build for Houston AI buyers

• Self-hosted LLM systems. Llama 3 / Llama 4 / Mistral / Qwen fine-tuned on the buyer\'s corpus, served via llama.cpp, vLLM, or Triton inside the buyer\'s infrastructure.
• RAG pipelines. Vector stores (pgvector, Qdrant), citation-first answers, versioned source documents, audit trails for healthcare and energy operations.
• Document-extraction workflows. Energy contractor documents, healthcare claim forms, insurance correspondence — structured data out of unstructured inputs.
• Internal copilots. Operations-team knowledge surfaces, customer-service triage, ticket categorization, on-call runbooks turned interactive.
• Eval + monitoring tooling. The harness that proves the model is good enough on day one; the monitoring that proves it stays good in week 26.

Houston verticals I work in

• Texas Medical Center healthcare-adjacent. Scheduling, claim review, audit logging, contractor portals, internal triage — not directly clinical work, which belongs to FDA-cleared specialists.
• Energy operations. Exploration data review, contractor-document extraction, predictive maintenance pipelines, operations dashboards.
• Microsoft-shop enterprises. AI features inside the Azure AD / Entra ID perimeter, Power BI-adjacent dashboards, Microsoft Graph integrations.
• Mid-market accounting + B2B services. Customer-service AI, document-extraction, internal copilots for the operations team.

How a Houston AI engagement is scoped

Sprint zero is the eval harness — the golden set and the measurable success criterion before any code ships. Sprint one is the first model. Sprint two is the production deploy. Sprint three is the monitoring. Total engagement typically runs 3-6 months. In-person discovery in the Houston metro is available for serious buyers.