Ardhendu Pramanik

CareerLens is a full-stack RAG (Retrieval-Augmented Generation) application I built from scratch to solve a real problem in my own job search — analyzing how well a resume matches a job description. I built the entire RAG pipeline without framework abstractions: paragraph-based chunking, vector embeddings via Google Gemini, and pgvector cosine-similarity search to retrieve the most relevant resume sections for any job description. The FastAPI backend uses proper service-layer architecture with Pydantic validation, SQLAlchemy ORM, and pytest coverage, with defensive error handling around LLM output parsing. The Next.js 16 frontend uses Server Actions and React 19, including a custom animated SVG score visualization. I deployed the full stack to production — Render, Vercel, and Supabase — and solved real infrastructure issues along the way, including IPv4/IPv6 connection pooling between Render and Supabase.

A full-stack AI-native collaboration platform built around a CRDT-backed multiplayer canvas (Liveblocks, React Flow) — conflict-free distributed state, live cursors, and presence that stay correct under concurrent edits. On the AI side, agentic workflows over OpenRouter turn natural-language prompts into live architecture diagrams using structured outputs and tool calling, constrained by a schema-validated action DSL with server-side sanitization so graph state stays consistent regardless of what the model returns. Long-running LLM jobs run on Trigger.dev with retries and SHA-256 idempotency keys for exactly-once execution. The stack spans Next.js 16, React 19, TypeScript, Prisma, PostgreSQL, Vercel Blob, and Clerk auth across 11 secured API routes. Built following the JS Mastery curriculum for structure — the goal was to understand every system underneath it, and I can walk through the design decisions and tradeoffs at each layer.

The dispatch module covered shipment assignment, driver scheduling, and status tracking across a live multi-team logistics platform. Dynamic Zod-validated forms, a scalable Prisma + Supabase data flow with production-grade handling of concurrency and data integrity, role-based access, and reusable form/table primitives.

Responsive, accessible UI for the public platform of the Indian Venture and Alternate Capital Association — an industry body serving 250+ PE, VC, and alternative-investment funds. Members, events, and research sections with optimized data-fetching (React Query) and Formik-driven forms, plus integration tests.