AI Engineer · Agentic Systems · Production LLM Pipelines
I build agents for a living. They still hallucinate. We're working on it.
I work at the intersection of LLM orchestration and production engineering — building systems where agents actually reason, retrieve, and recover from their own mistakes, then shipping them somewhere real. My projects aren't independent experiments; they follow one question: how do you make an AI system you'd actually trust in production?
| Project | The Question I Was Answering | Result |
|---|---|---|
| Apollo | Can a multi-agent system reason across clinical data without hallucinating on doctors? | Parallel agents, hybrid RAG, eval agent that catches bad answers before they surface |
| Fraud Detection | Can you score 600 transactions/sec in real time and still catch subtle fraud patterns? | Kafka + LightGBM pipeline, Redis feature store, PR-AUC ~0.94, +15% precision |
| E-Commerce Warehouse | Can one person build a full cloud data warehouse with CI/CD and a live dashboard? | AWS S3 + Redshift + Airflow + GitHub Actions + Plotly Dash — solo, 0 to 1 |
AI Backend Engineer · PM Accelerator
Built a production agentic AI platform from scratch — LLM-powered retrieval pipeline on OpenAI embeddings and MongoDB Atlas, 90% query resolution accuracy across 10K+ live queries. Designed a classification + fact-verification layer with guardrails hitting 85%+ confidence end-to-end. Led a 4-person team.
Key decisions: embedding model selection, chunking strategy, confidence threshold tuning, guardrail design.
⚡ open to AI engineer & ML roles — full-time or founding team