I build web apps, backends, AI infra and more, from idea to launch.
12 agents know his work. Ask them anything.
Every answer comes from my own notes and is fact-checked before it reaches you.
Three years, five teams, founding engineer at two of them.
Finance for India's creators and freelancers, run by AI agents. Invoices, international payments, GST filing, taxes, expenses and more all happen from one chat, and I built the agents behind it.
TypeScript
Google ADK
MCP
Xflow
Cashfree
GCP
Node.js
React
PostgreSQL
LangChainAAG-UISSSE
Cloud Run
DockerKZerodha Kite
Google Calendar
YouTube
OAuth 2.1I built the backend for a Canadian marketplace, its services, its payments and the events that connect them.
Node.js
Pub/Sub
Stripe
Redis
MongoDBDocker
Bitbucket
JWTAI news intelligence for journalists. I owned the Python backend, end to end.
Python
FlaskLangChain
OpenAI API
Pinecone
Firebase
DynamoDBDockerAn enterprise analytics product I built from the first commit, as its founding engineer.
Next.jsTypeScriptDynamoDBLangChainOpenAI APIGCPDockerUUmamiMy first role, in a four-person cross-functional team. The real-time notifications I built lifted engagement 30%.
Next.jsFirebase
Cloud FunctionsNine side projects, all still running. Ask about any one and an agent will walk you through it.
Twelve agents that know my work. Ask a question and the orchestrator sends the right ones; they answer from my notes and GitHub, and a critic checks every claim.
One OpenAI-compatible endpoint in front of OpenAI, Claude, Gemini and Groq, with automatic failover, circuit breakers, per-tenant budgets and a 24-hour cache.
A robot that streams its camera to Gemini and turns the model's reasoning into motion, with no maps and no hand-coded vision. Every command passes a safety check first.
A live world map of memes. When a coordinated raid hit it, I shipped moderation and rate limits under fire, then relaunched it GIPHY-only.
A four-joint arm in simulation that plans its own collision-free path with MoveIt2 and inverse kinematics.
Runs code at scale on plain AWS: jobs wait in an SQS queue, EC2 workers pick them up, and results land in DynamoDB.
Teachers turn a PDF, a page of text or a link into an assignment in seconds, and GPT grades the submissions.
A small platform-as-a-service: one command takes a GitHub URL, builds the frontend, containerises the backend and puts Nginx in front.
Turns a whole directory into one LLM-friendly context file, so a codebase fits in a single paste.
Student projects from before the jobs: small tools that taught me to ship.
GitHub
Write-up