I build AI agents, memory systems and developer tools, and write practical guides about how they work.
Portfolio · Writing · LinkedIn · Resume · Email
Open to technical writing and developer contracts: tutorials, API demos, documentation and agent-tool integrations.
| Project | What it does | Explore |
|---|---|---|
| Flatpack | Turns narrated how-to videos into printable illustrated instructions, with a review step before printing. Local prototype using Livepeer Agent and ffmpeg. | Demo |
| ₹800 | Plans outdoor work around configured heat rules using OR-Tools. Shows the income shortfall when jobs cannot fit. English and Hindi interfaces. | Demo |
| otel-swarm | OpenTelemetry instrumentation for agent workflows, model fallbacks and review events, with SigNoz dashboards and alerts. | Technical guide |
| ContextFirewall | MCP prototype that checks agent memory for stale facts, contradictions, secrets and missing evidence. Built with Cognee, FastAPI and Next.js. | Architecture and setup |
| OmniDev | Native macOS assistant for searching files and screenshots with citations, editing code and running tests. Local Ollama inference with an optional cloud provider. | Recorded demos |
| Pixel Digit Recognizer | Runs a NumPy digit-recognition model in the browser through Pyodide, with no inference backend. | Try it |
Flatpack placed 2nd in the Agent Builder track at the Livepeer Agent Hackathon 2026.
- ToolTrap: testing what agents repeat from untrusted tool results. A benchmark write-up covering source boundaries, scorer corrections and experimental limitations.
- The smolagents bug that made my agent retry the same valid code three times. A debugging walkthrough recognized as a DEV Summer Bug Smash runner-up.
- Tracing a multi-agent LLM system. A practical guide to instrumenting agents with OpenTelemetry and SigNoz.
Python · TypeScript / JavaScript · FastAPI · React / Next.js · SQL · Docker · AWS · MCP · OpenTelemetry