Agents in production,
and the ones that taught me how.
A working catalog. Some are live with users; some are research; some are old work I keep around to remember where I started. Filter by stack to find what's relevant.
DocIQ
Multi-tenant document intelligence. Agentic RAG over enterprise contracts, invoices, technical specs. Hybrid search, structured extraction, citation grounding, audit trails.
OMS-v1 Copilot
Order management copilot. Natural-language operator agent: queries inventory, flags exceptions, drafts vendor emails, schedules dispatch. Replaces 60% of ops handoffs.
AI Battlefield
Adversarial multi-agent simulator. Two agent armies negotiate, deceive, and cooperate over a shared world model. Tests alignment failures under self-play pressure.
ERP-v1 Layer
Agentic interface bolted onto a legacy ERP. Conversational queries, audit-grade actions with confirmations, role-aware tool routing. ERP becomes typeable, not clickable.
RAGs toolkit
A small, opinionated retrieval-augmented generation toolkit. Reusable chunking, hybrid search, and ranker primitives I keep reaching for across projects.
Hyderabad Rent
Geospatial rental search with natural-language filters. Agent translates "2BHK near metro under 30k" into structured queries against a city-wide listings index.
Drone Computer Vision
Mediapipe + DJI Tello — real-time detection of unsafe behaviors from a mini drone. Demoed at Microsoft Reactor and IoT shows. The project that started everything.
Stock LSTM
Sequential prediction over historical equity prices using LSTMs. Quaint by 2026 standards, but the discipline of getting features right still applies.
Multilingual Doc Classifier
Document classifier across 12 languages using transformer encoders. Held up well in production for years before LLMs ate this category.
DiRA · Web3 Health Records
Digital Health Records on a permissioned ledger. Patient-controlled access, audit logs, and selective disclosure. Speculative; informed how I think about agent permissions today.
Socialy
Re-implemented LinkedIn-style connection-degree algorithm to learn graph traversal at small scale. Dijkstra plus heuristics.
Travel Place Recommender
Group-conflict resolver: maximize aggregate utility across a friends-group's preferences. A reminder that "AI" was group-decision-theory before it became LLMs.
Curious about how an agent got built?
Happy to walk through architecture, trade-offs, or what didn't work. Drop a note.