Available · AI Engineer
I build AI systems that see, retrieve, reason & ship.
CS master's graduate from Indiana University focused on agentic pipelines, multimodal retrieval, production ML, and cloud-backed systems that hold up outside the notebook.
A retrieval system that answers questions over visual frames and spoken transcript, then returns timestamped evidence you can jump back to.
A multi-agent surveillance system: one agent watches every frame, one reasons about what's worth an alert, one turns plain-English commands into monitoring rules, and one keeps searchable memory of everything it has seen.
A campus room-booking platform where students, faculty, and staff browse an interactive floor plan and reserve rooms by role, with per-role limits enforced at the API gateway. An event-driven microservice backend communicates asynchronously over RabbitMQ so the booking flow stays resilient under load.
An LLM travel planner that turns a plain-English request into a day-by-day itinerary, packing list, and budget. A guardrail-fronted agent extracts trip features, fans out to five live data sources in parallel, then composes the plan through a provider-agnostic LLM layer with automatic fallback — backed by a judge-LLM harness that scores plan quality across five metrics.
Every project on this page shipped past the demo. That's a different bar than it sounds like.
A retrieval system gets graded against held-out queries before it gets a UI. An agent pipeline gets a fallback for every model call, because APIs go down, get slow, or return garbage. And nothing ships until I've pushed it past the load it was built for, since that's usually where the real bugs hide.
None of it shows up in a screenshot. It's the difference between a project that looks good in a demo and one that survives a Tuesday in production.
A demo proves it can work once. An eval set proves it works enough. I build the harness first.
APIs go down, get slow, or return garbage. Each one needs its own plan before the system ships.
Bugs at 10 requests a second don't look like bugs at 1. I find that out before a user does.
A LangGraph AI companion built to feel like texting a real person. Ava remembers what you tell her, knows what she's doing right now, and replies in text, voice notes, or images. Built from scratch with Mouryan Jayashankar and tested against live APIs. Tell me what it should do next.
Architecture and setup are in the repo.
Future Wonder Co.
MyEdMaster LLC, Leesburg, VA
Fox Trading (1STOP.AI), Remote
Infosol Technosol, Remote
Indiana University Bloomington
GPA 3.80Jain University
GPA 8.97 / 10IU Luddy Hackathon 2026
Dynamic Leaderboard Case. A real-time leaderboard that ranks Indianapolis districts by infrastructure health.
Luddy InnoQuest
AI-Powered Video Intelligence Platform, with a four-person team.
§ 07 · Contact
Hiring, have a lead, or want to talk agentic systems, multimodal RAG, or production ML? My inbox is open.