I build AI systems end-to-end β model architecture, retrieval pipelines, backend infrastructure, and the production UI that ships them. My focus is agentic AI, RAG architectures, neural retrieval, and the backend engineering that turns a model into something people can actually run.
Four independent products shipped so far: a local-first desktop AI agent, a neural recommendation engine, an LLM safety-rails prototype for care-ops triage, and an AI analyst workspace for financial data β each one taken from architecture diagram to working release, not left as a notebook.
I hold an MS in Management Information Systems from UT Dallas and a BE in Computer Engineering from the University of Pune.
Right now I'm focused on:
- π₯οΈ Building LocalMind β a privacy-first, local-only desktop AI agent (Tauri + Rust + Ollama)
- π§ Designing agentic pipelines and retrieval architectures (RAG, HyDE, two-tower recommenders)
- βοΈ Backend systems for ML products β FastAPI services, vector search infra, production deployment
Languages
AI / ML
Frameworks & Frontend
Data, Cloud & DevOps
Tools & Platforms
|
π₯οΈ LocalMind Local-first desktop AI agent (Tauri/Rust + React/TS + Python/FastAPI sidecar) with hierarchical memory, HyDE retrieval, and a local knowledge base over Ollama + LanceDB. No cloud, no accounts, nothing leaves the machine. |
π¬ Kino Two-tower neural retrieval system (PyTorch + FAISS) with an MLP ranking layer and IPS-weighted evaluation, served via FastAPI behind a Netflix-style Next.js frontend. |
|
π©Ί TriageCopilot An event-sourced care-ops prototype where an LLM agent triages synthetic patient intakes with tool-calling, retries, and enforced safety rails β every run ends in a mandatory human handoff, by design. |
πΉ LedgerLens An AI analyst workspace for SEC filings and macro data, with RAG over Ollama, SSE streaming, and a FastAPI + Next.js stack. |
