I build AI systems that ship, scale, and pay for themselves.

A consumer-research platform that reads beyond liking scores — capturing explicit and implicit responses to aroma and food products. Designed, built, and operated end-to-end: Django, React and PostgreSQL on AWS ECS with Cognito auth and hands-free CI/CD — regional teams get updates with zero manual server work.

• Django
• React
• PostgreSQL
• Docker
• AWS ECS
• GitHub Actions

A multi-agent reasoning system that lets people talk to their tools. LangGraph agents orchestrate MCP servers and RAG over internal documentation and vector databases — structured outputs, evals in LangSmith, served over FastAPI.

• LangChain
• LangGraph
• MCP
• RAG
• Qdrant
• LangSmith
• FastAPI

Interactive formulation optimization for flavorists. High-throughput FastAPI pipelines run SciPy linear-programming solvers and stream results back into the React UI in real time as they converge.

• React
• FastAPI
• SciPy
• Linear Programming

An immersive aroma-evaluation environment built in A-Frame — 4K 360° video, adaptive resolution, and interactive 3D controls wired into external sensory hardware. Locally sideloaded for deterministic, network-free lab trials.

• WebXR
• A-Frame
• JavaScript

A private LLM serving stack on an NVIDIA DGX Spark — vLLM and Ollama behind a LiteLLM gateway and Caddy reverse proxy, with API-key management and Postgres usage tracking. Full GPU observability via Prometheus, Grafana and DCGM, Redis-backed scheduled agents, and a RAG pipeline over SearXNG metasearch and Qdrant.

• vLLM
• LiteLLM
• Caddy
• Qdrant
• SearXNG
• Prometheus
• Grafana