AI Reliability Strategist — I build the systems that work while you don't.
Most LLM demos look impressive until they meet real users, messy data, edge cases, latency, cost limits, weak retrieval, and broken workflows. The model call works. The system around it collapses.
I help founders and small teams turn fragile AI prototypes into production-ready systems:
- RAG pipelines that retrieve the right context and cite sources
- Agent workflows with clear tool use, failure handling, and human override
- Guardrails that reduce hallucinations, compliance mistakes, and operational risk
- Evaluation paths, observability, and audit trails
- Clean architecture and handoff docs your team can own
Five-service monorepo for WH-347 federal payroll compliance. React 19 · Vercel AI SDK · FastAPI × 2. 271 tests, 0 failures. Every compliance decision cites the statute.
Next.js dashboard for tracking LLM costs, latency, and cache savings by workflow. Built for teams running multiple AI providers who need visibility into what's actually expensive.
Trace multi-step AI agent executions, log tool calls, measure latency, and debug failures in production. Built for LangGraph/Mastra agent stacks.
Automated recordkeeping, regulatory checks, and audit trail generation for US federal payroll and labor law.
| Repo | Description |
|---|---|
| WCP-Compliance-Agent-V5 | Five-service monorepo for WH-347 payroll compliance · 271 tests · React 19 · FastAPI × 2 |
| compliancelens | Compliance automation toolkit — recordkeeping, regulatory checks, audit trails |
| costpilot | LLM cost, latency, and savings dashboard — Next.js + Python + PostgreSQL |
| agenttrace | Agent tracing and observability tooling for production AI |
| super-study | AI learning repo — RAG experiments, agent architectures, LLM evaluation research |
| palindrome-checker | Study project — JS test suite, CI/CD learning |
| FishRaposo | This profile — AI Reliability Strategist · Production RAG · agent workflows |
Python TypeScript FastAPI OpenAI Mastra LangChain LangGraph RAG PostgreSQL pgvector
- WCP V5 — Five-service monorepo · 271 tests · every decision cites the statute
- 7 production AI systems shipped at Expat Money (GPT assistants, RAG workflows, scraping pipelines, automation tools)
- 271 tests, 0 failures across the WCP V5 monorepo
- Portfolio — case studies and live proof
- Upwork — hire me for AI reliability work
- LinkedIn — AI reliability thinking
I don't sell chatbot wrappers. I build the infrastructure underneath them.