Systems I build around AI

I don't just call model APIs — I design the graph around them: who plans, who checks, when to loop, and what changes when the user's context changes. Three systems, each one explorable node by node.

• Adaptive multi-agent graph

  NeuroFlow →

  A brain-computer-interface loop: EEG decodes the user's cognitive workload, and the LangGraph agent topology reshapes itself around it. Switch the workload to watch the route change.

  13 nodes · 15 edges · 3 adaptive states

• Agentic code-generation workflow

  Traycer-mini →

  Plan, generate, review, approve, verify. AI writes the code — but nothing reaches the codebase without a reviewable plan, a diff and a passing check.

  9 nodes · 10 edges

• Research · NIT Bhopal

  HTR Bench →

  Ongoing research on vision-language models for handwritten text recognition in multilingual and Indian-language documents — benchmarking OCR and VLM systems head-to-head.

  7 nodes · 9 edges