Things I built to find out how they work.

An AI-native desktop IDE whose agents read your repo, run your tests, and teach from what you actually got wrong.

• Mastra research + mentor agents that inspect repositories, execute tests, and review diffs before teaching anything
• Persistent learner memory so personalization survives across projects instead of resetting each session
• Per-feature model routing with Phoenix tracing, keeping agent runs cost-aware and actually debuggable
• Turborepo monorepo spanning the Electron shell, agent runtime, and Prisma/Postgres persistence layer

An adaptive coding gym that turns your attempt history into the next challenge you should face.

• Ability Ledger converts attempt history into personalized next challenges instead of a fixed problem list
• Isolated sandboxed runners execute untrusted submissions across three languages
• Persistent sessions so a practice run survives reloads and resumes mid-problem

An AI roadmap builder that turns a fuzzy goal into an execution map with milestones and dependencies.

• Goal intake that turns rough intent into a structured, guided roadmap request
• React Flow execution canvas with nodes, dependencies, time estimates, and detail panels
• Plans, auth, billing, and loading-state polish that push it from demo energy to product

Five blockchain layers written from scratch in pure Python, with zero external dependencies.

• Proof-of-work mining and Merkle validation written from first principles, zero dependencies
• Gossip-based P2P networking with a bootstrap server keeping peers on the longest chain
• JSON-RPC node interface plus tamper detection and chain-recovery paths for bad state

The AI career-guidance platform that won Smart India Hackathon 2024 against 80,000+ teams.

• Landing experience that makes counselling, AI roadmaps, and interview prep legible in one screen
• Student dashboard with lesson progress, skill tracking, and guided navigation
• Owned the product experience end to end under hackathon pressure, and won nationally

A React Native group-settlement app with UPI checks, voice input, and OCR receipt parsing.

• Owned the full stack: RN client, FastAPI + SQLite backend, and Expo Cloud CI/CD
• AI voice input and OCR receipt parsing cut split creation time by about 45%
• Payment verification checks around UPI settlement flows