DataDriven

DataDriven started because I kept failing data engineering interviews. Not because I couldn't write SQL, but because practice platforms tested the wrong things. LeetCode tests algorithms. Real interviews test whether you can design a schema under pressure, debug a window function with an interviewer watching, or explain why your pipeline would break at scale. So I built what I wished existed: a platform where you practice the actual interview, not just the code. Every challenge is sourced from real interviews at real companies. The AI interviewer pushes back like a real one. Your code runs against real data. No multiple choice, no h

Next.js 14 (App Router) on EC2 with a Python FastAPI backend serving 223 endpoints. PostgreSQL on RDS. Real SQL execution via sql.js in the browser. Python runs in Docker sandboxes with 4 parallel workers. Client-side AI interviewer powered by WebLLM (Qwen3 4B) running entirely on the user's GPU, no server round-trips for conversation. OpenAPI schema auto-generates TypeScript types from Pydantic models. SWR for data fetching. Adaptive learning uses IRT, FSRS memory decay, and a custom GRU-based skill model. Canvas-based schema and pipeline architecture editors built from scratch. Deployed via rsync to EC2.