Software engineer building AI agents, developer tooling, and polished product UI, from research papers to shipped code.
tweenly is an open-source library of 60 animated React components for shadcn/ui, distributed as a shadcn registry. I designed and built all of it: the components, the docs site, and the tooling that ships them.
- Live playground for every component. Every prop is a control, and the usage code rewrites itself as you tweak it, so you copy exactly what you tuned.
- Charts that take real data. Nine chart types share one chart kit (scales, formatters, tooltips, legends) and accept plain row data, editable live in the docs.
- Page-level motion that works with the Next.js App Router. Route transitions, page loaders, sticky-reveal footers, scroll storytelling and physics-based canvas confetti.
- Built for adoption. One-command install through the shadcn CLI, a guided product tour, per-page share images,
llms.txtfor AI assistants, and analytics on installs.
Live site · Source · Changelog
| Project | What it is |
|---|---|
| IntentGate | An MCP gateway that makes AI agents stick to what they said they'd do. The agent declares a plan, ArmorIQ signs it, and every tool call is checked against it, so a prompt-injected agent can't read your keys or post them anywhere. Works with any MCP client and ElevenLabs voice agents, with a Flight Recorder to replay every decision. Built in a week for the ArmorIQ × ElevenLabs hackathon. Live demo TypeScript · MCP · ArmorIQ · ElevenLabs |
| vaak | Self-hosted, open-source speech-to-text for English, Hindi and other Indic languages: one Python API, CLI and server over six open-weight backends (Whisper, IndicConformer, Parakeet, wav2vec2, Vosk). Includes a benchmark harness where every number says where it came from, and a 14-part tutorial on how speech recognition works. Docs Python · Whisper · FastAPI |
| Argus | An AI agent that investigates incidents, built for the SigNoz hackathon. It correlates telemetry, logs and traces with memory to reason about the root cause, instead of just surfacing raw signals. Python · LangGraph · OpenTelemetry |
| Tiny Stories | A ~30M-parameter GPT-style language model trained from scratch on TinyStories. I wrote the transformer and training loop to understand how LLMs work beneath the API layer, then audited my own run and wrote up what I'd fix. Python · PyTorch |
| Bookeeda | A startup I built in college, taken from idea through user research to product. Shortlisted for PIERC's incubation program from 4,500+ applicants. |
| Linkerly | A workspace for organizing, sharing and analyzing URLs. Pre-launch. |
| Role | Where |
|---|---|
| DevOps Engineer | Onwe: infrastructure and engineering work across the stack |
| Technical Content Writer | SigNoz: developer-facing technical content for a YC-backed observability company |
| DevOps & Open Source Lead | Google Developer Student Club: ran technical sessions on developer tooling for students |
AI / ML: PyTorch, scikit-learn, Pandas, NumPy, LangChain, LangGraph, Hugging Face, OpenAI and Anthropic APIs Frontend: React, Next.js, TypeScript, Tailwind CSS, shadcn/ui, Motion, GSAP Backend and infra: Node.js, Express, FastAPI, PostgreSQL, MongoDB, Redis, Docker, AWS, GCP, Linux Observability: OpenTelemetry, SigNoz, Grafana, Prometheus, Jaeger
Currently exploring: AI agents, RAG, distributed systems and developer tooling. Open to: software engineering roles in AI, developer tools and product engineering. Say hello.