Role
AI Engineer · Tech Lead
Problem
Enterprise customers like Mistral and Apple needed a predictable path from LLM experimentation to compliant, observable production rollouts. Each deployment required domain-specific grounding data, synthetic fixtures that mirrored third-party APIs, and guardrails to keep autonomous agents from overstepping permissions.
Systems approach
- Designed RLHF and SFT loops with human-in-the-loop review and automated drift detection.
- Built RAG pipelines that layered vector search with policy-aware prompt orchestration.
- Cloned Slack, Amazon, Jira, and PayPal APIs in FastAPI to stress-test agent reasoning before go-live.
- Configured MCP servers so AI agents could safely choose the minimal endpoint set per task.
Automation & tooling
- Python + SQL pipelines generated synthetic partner data, boosting data prep throughput by 30%.
- Observability hooks piped metrics into Grafana to track hallucinations, latency, and cost per token.
- Code review checklist codified safety tests, enabling a distributed team of five engineers to ship in sync.
Results
Faster data engineering cycles
Enterprise clients onboarded
Production regressions over 12 months
What I owned
Architecture
Authored the reference architecture for multi-tenant RLHF pipelines, privacy-aware embeddings, and MCP routing. Documented upgrade paths and cost maps for leadership.
Team lead
Led five engineers, ran design reviews, paired on tricky FastAPI services, and instituted code quality bars that held across time zones.
Safety & governance
Embedded automated policy checks, audit logging, and data synthetic monitoring so compliance teams could trust each deployment blueprint.