I'm a full-stack developer with an Electronics & Communications Engineering background — self-taught in software and AI systems. I build and ship real products end-to-end: the app, the backend, and the AI features layered on top. My main project, DevCastle, is a live developer community platform with real users that I've built and run solo.

I care about shipping things that actually work for real users, not just demos — which means paying attention to type safety, caching, and what happens when an LLM call fails or times out.

A community platform for developers: discussion feeds, project launches, a job board, and an AI assistant — built and maintained solo.

Live: devcastle.vercel.app  |  Scale: real users tested (currently in making)· ~174K lines of TypeScript · 25+ database models

• AI Assistant — In-app assistant using Groq for inference, local embeddings (@xenova/transformers), and Upstash Vector for retrieval-augmented answers about the platform. Built in stages: streaming chat → tool calling against the database → RAG → multi-agent coordination via LangGraph → rate limiting.
• Reddit Market Intelligence — Scrapes subreddit discussions (via Decodo + Cheerio) and uses an LLM to score and summarize SaaS opportunities from them — demand signals, existing competitors, monetization ideas.
• Community & Social Feed — Sub-communities, threaded comments, an editor for long-form posts, project launches with voting.
• Job Board — Full-text search over listings with filters for role, compensation, and stack.
• CI/CD — GitHub Actions pipeline running lint, typecheck, Lighthouse, CodeQL, and secret scanning (TruffleHog) on every push.
• Analytics — PostHog integrated for product usage tracking.

Stack: Next.js 14 (App Router) · TypeScript · MySQL + Prisma · Upstash Redis/Vector · Groq · LangGraph · Vercel

A travel planning tool that splits the problem across a few cooperating agents instead of one big prompt. GitHub Repo

• Built with LangGraph, running specialized agents (flight search, hotel search, itinerary synthesis) in parallel with shared state.
• Session state persisted in PostgreSQL so a planning session can be resumed.
• Streamlit frontend for interacting with the planner.

Stack: Python · LangGraph · LangChain · Streamlit · PostgreSQL

Languages: TypeScript · JavaScript · Python · SQL

Frontend: Next.js · React · Tailwind CSS · Streamlit · TanStack Query · Zustand

Backend: Node.js · Express · FastAPI · tRPC · REST/GraphQL · Zod

Data: PostgreSQL · MySQL · MongoDB · Prisma · Redis · Pinecone

AI/LLM: LangGraph · LangChain · OpenAI API · Anthropic Claude API · Google Gemini · Groq · RAG (dense/hybrid retrieval)

Infra: AWS · Docker · Vercel · GitHub Actions · Sentry

Open to full-stack and AI engineering roles, and to conversations about AI product work.

AWS Certified: AWS Cloud Technical Essentials — 98%