AI / ML Engineer · Agentic Systems
// I build AI agents that remember, reason, and ship.
Multi-agent systems, memory and retrieval infrastructure, and applied LLMs — taken from hackathon stage to real users.
About
I'm an AI/ML engineer focused on agentic systems: multi-agent orchestration, memory and retrieval infrastructure, evaluation, and getting LLMs to behave reliably in production. I just finished my M.S. in Applied Data Science at San Jose State, and I've spent the last two years building, from production LLM systems at a startup to a long list of hackathon projects that actually ship.
I like the full arc: training and fine-tuning the model, building the system around it, and putting it in front of real users. Along the way I've taught hundreds of students to use AI through the OpenAI campus program, because the work isn't done until someone else can run it too.
Selected work
A mix of a startup, award-winning hackathon builds, and research projects. Award badges are real placements.
A multi-tenant company-brain that ingests a team's tools into a shared memory store and serves it back to people and AI agents, with Notion OAuth ingestion, streaming RAG chat, per-principal access control, and an MCP server for external agents.
View →An NLP pipeline that parses research PDFs, extracts entities and citations into a Neo4j knowledge graph, and answers questions with graph-aware retrieval grounded in evidence across many papers.
View →A 6-agent medical-navigation assistant that turns spoken symptoms into structured guidance and matched specialists in under a minute, with an independent verifier that reviews every output for unsafe answers before it reaches a user.
View →A multi-agent system for ad creative and strategy, built on Gemini 2.0, that took the top prize at the Adsystems hackathon.
View →A 24/7 voice agent that answers a city's 311 line, files real service tickets, and routes life-safety calls to 911 — with a real-time React + FastAPI operator dashboard over server-sent events, deployed on a live phone number.
View →A production-style customer-support agent on a checkpointed, idempotent pipeline that indexes a 1.4M-row catalog into Qdrant, with native LLM tool-calling over live PostgreSQL and sentiment-driven escalation.
View →More work
Llama 3.2 3B enterprise router and a Llama 3 8B support model, fine-tuned and 4-bit quantized for efficient serving.
PyTorch decoder mapping 512-feature neural signals to text, tuned against word error rate.
Security investigation agent on Claude 3.5 with a Neo4j graph and a Next.js front end.
Health companion built with ElevenLabs, Terra, and LumaLabs.
7-agent web workspace that turns raw product ideas into launch-ready assets, with teams and RBAC via Stack Auth.
Agent that drafts meeting prep in Notion from Gmail context.
GPT-style transformer built in PyTorch: pretraining, SFT, LoRA, and DPO, end to end.
ONNX model quantized and deployed to an NPU, benchmarked on latency and PSNR.
Vision-based extraction of covenant terms from financial PDFs with real-time breach detection.
Experience
Leadership & AI enablement
Skills
Education & research
San Jose State University · Aug 2024 — May 2026
Coursework: Machine Learning, Deep Learning, Generative Models, NLP, Computer Vision, Distributed Systems.
Gujarat Technological University · Aug 2019 — May 2023
Research: LLM inference-latency optimization · Visual Speech Recognition (vision + speech)
I'm looking for AI/ML, applied-AI, and forward-deployed engineering roles. The fastest way to reach me is email.