AI Interview Platform
AI-powered async interview platform — parses job descriptions, generates rubric-backed questions, runs text, voice, and coding interviews, then scores answers and produces reports with integrity signals.
//The Problem
Screening interviews consume enormous engineering and recruiter time, and take-home tests are inconsistent and easy to game. Companies need a way to evaluate candidates asynchronously without sacrificing rigor or fairness.
//My Role
Solo builder — product design, architecture, full-stack implementation, and billing.
//Key Decisions & Work
- Designed a two-service architecture: a Next.js 15 web app and a FastAPI AI service with RQ background workers, keeping LLM workloads isolated from the user-facing app.
- Built a pluggable LLM router that switches between providers (Groq, Ollama, OpenAI, Anthropic) without touching feature code — controlling cost while avoiding vendor lock-in.
- Implemented JD parsing that generates rubric-backed questions, with scoring that returns structured reports including integrity signals.
- Added text, voice, and live coding interview modes — Monaco editor with self-hosted sandboxed code execution.
- Wired subscription billing with checkout and a customer self-service portal.
//Impact
- Functional end-to-end: JD in → interview → scored report out, with semantic search over answers via pgvector.
- Deliberately accessibility-aware design: per-question timers are soft limits rather than auto-submission.