Projects

These are products, not notebooks. Each one solves a problem I actually had — grounding answers in real documents, automating an application pipeline, turning a resume into a website — and each is built with the evaluation, guardrails, and observability that make an AI system trustworthy enough to ship.

4

Shipped to production

3

Live deployments

4

Open source repos

18

Technologies used

RAG Platform
~/corpuspilot01

CorpusPilot

Multi-domain RAG platform for grounded document Q&A

Upload documents across five domains — legal, financial, healthcare, research, technical — and ask questions answered strictly from retrieved chunks, with mandatory inline citations.

  • Hybrid semantic + lexical retrieval with domain-weighted Reciprocal Rank Fusion and Cohere reranking
  • SQL-RAG pipeline that turns financial filings into safety-filtered SQL for precise numeric answers
  • RAGAS evaluation gate wired into CI — merges that regress faithfulness are blocked automatically
FastAPIReactLangGraphLlamaIndexMilvusPostgreSQLRAGASLangSmithPrometheus
Live DemoGitHub
Multi-Agent Platform
~/applyai02

ApplyAI

11-agent pipeline that automates the entire job application

Drop in a resume and a job posting; an orchestrated chain of agents parses both, runs gap analysis, and hands back ATS-optimized application materials — tailored, scored, and ready to send.

  • 11-agent LLM orchestration built on Google ADK + LiteLLM, traced end-to-end with LangSmith
  • Anti-hallucination guardrails restrict tailoring to skills verifiably present in the real resume
  • Hybrid deterministic + LLM ATS scoring drives an automated rewrite feedback loop
FastAPINext.jsGoogle ADKLlamaIndexGroqOpenAIGCPLangSmith
Live DemoGitHub
AI SaaS Product
~/auracv03

AuraCV

Résumé in. Portfolio out. Twenty seconds.

An open-source platform that turns a resume PDF into a polished, hosted portfolio on a personal subdomain — no forms, no drag-and-drop builders. Live at auracv.me with real users.

  • LLM extraction pipeline parses any resume PDF and typesets it into a designed portfolio
  • Hallucination-safe extraction — nothing appears on the page that isn't in the source document
  • OAuth sign-in, Supabase row-level security, and instant publishing to yourname.auracv.me
Next.jsFastAPISupabaseLlamaIndexOpenAI
Live DemoGitHub
Autonomous Agent
~/meesho-coder04

Meesho Coder

A Claude Code–style autonomous coding agent

A from-scratch coding agent that takes a plain-English brief and writes the project itself: a LangGraph state machine plans, architects, then loops a coder node until the build is complete.

  • Planner → architect → coder graph with conditional looping until the task reports complete
  • Sandboxed file tools with path-traversal guards — the agent can only write inside its project root
  • YAML-configured prompts per node, Groq-served models, and a pytest suite over the graph
PythonLangGraphLangChainGroqPydantic
GitHub

Want to see more?

Most of the work above is open source — read the code, or get in touch if you'd like to talk through how any of it was built.

Browse GitHubGet in Touch