I engineer intelligent systems. From local semantic search engines to full-stack streaming chat architectures, I build the infrastructure, APIs, and interfaces that take AI models from research to production.
A semantic code search MCP server with local embeddings, hybrid retrieval, and WAL-backed persistence.
A closer look at the architecture, trade-offs, and engineering outcomes of key systems I've built. For smaller experiments, visit my GitHub.
A local MCP server that indexes codebases into embeddings and exposes hybrid search to AI coding assistants
A full-stack AI chat system with streaming responses, provider switching, and real-time sentiment analysis
A provider-agnostic LLM gateway unifying five AI providers behind a single Next.js API layer
I've taken models from research notebooks to production inference serving real users. That means the full path: PyTorch and Transformers for the model layer, FastAPI for serving, Cloudflare Workers for edge deployment, and the provider orchestration to make multi-LLM switching invisible to users.
When a problem needs a product rather than a pipeline, I build that too. TypeScript, Next.js, React, databases, auth, streaming — the frontend is not a handoff concern, it's part of the system I own. I've shipped browser extensions, real-time dashboards, and multi-provider chat interfaces.
FastAPI for async Python services, Node.js for JavaScript runtimes, Cloudflare Workers for edge compute. JWT auth, SSE streaming, rate limiting, provider-agnostic API layers — the boring infrastructure that makes products feel reliable.
Multi-provider LLM orchestration across OpenAI, Anthropic, Gemini, and xAI. Prompt engineering, sentiment analysis pipelines, semantic code search with local embeddings, and the MCP protocol infrastructure connecting AI to developer tools.
B.E. in Computer Science
Built and shipped end-to-end AI products across browser extensions, geospatial dashboards, multi-provider chat systems, and developer tooling — each solving a concrete user problem with real deployment infrastructure.
Full ownership across model integration, backend architecture, streaming protocols, and deployment to Cloudflare, Vercel, and Chrome Web Store.
I started building because I got frustrated by tools that almost worked. A Coursera quiz that took four hours of video to answer. A forest monitoring dashboard that required five separate browser tabs. A chatbot that couldn't tell you whether the user was happy or confused. Each frustration became a shipped product.
The thread connecting everything is the distance between a working model and a working product. Taking an LLM integration from "it generates text" to "it streams tokens through SSE, persists conversation state, switches providers without downtime, and scores sentiment in parallel" — that's where I spend my time.
I don't hand off. I build the model integration, the API layer, the streaming protocol, the React frontend, and the deployment pipeline. Not because I want to do everything, but because the seams between layers are where most products break. A backend engineer who doesn't understand streaming UX builds a system that feels laggy. A frontend developer who doesn't understand provider abstraction builds an interface that breaks when you switch from Gemini to OpenAI. I close those gaps.
Getting GPT-5.5 to generate good text takes an afternoon. Getting that text to stream reliably through SSE, switch providers without dropping a session, and degrade gracefully when the API returns a 429 — that takes weeks. The engineering that matters lives in the infrastructure around the model.
My forest monitoring dashboard has a "no-mock mode" — a toggle that refuses to display any data that isn't verified live. When an API is down, the user sees an honest error, not a plausible lie. This principle extends to everything I build. I don't claim numbers I can't source.
Right now I'm deep in the infrastructure connecting AI to developer tools — semantic code search with local embeddings, MCP protocol servers, and retrieval systems that work without sending your code to a cloud API. I'm looking for teams that ship AI products to real users, where the work matters past the demo stage. I hold a B.E. in Computer Science.
I'm looking for AI engineering roles where I own the full stack from model integration to deployment. If you're building something where the engineering problems are genuinely hard — I want to hear about it.