Full-Stack AI Engineer — Build an AI-Native Learning Platform (RAG + Next.js + Supabase)

# Full-Stack AI Engineer — Semantic Search + Next.js + Supabase (Long-Term, Contract-to-Hire) ## About We're building an AI-native platform that makes a large archive of recorded talks genuinely discoverable and useful: need-based semantic search over transcribed media, with a subscription product built around it. We have a clear product vision and architecture and are looking for a lead engineer to build the first version and grow with us long-term. Full product details are shared with shortlisted candidates under NDA — this post focuses on the engineering and the skills we need. ## The engineering challenge You'll build a two-part system that shares one database: 1. **A content pipeline (Python):** ingest recorded talks, transcribe them, chunk and enrich the transcripts with metadata using an LLM API, generate embeddings, and store everything in Postgres. 2. **A web app (Next.js):** fast, crawler-friendly, SEO-strong content pages with structured data; retrieval-based search that returns relevant source material with links/citations; user accounts; and Stripe-gated paid content. We care a lot about retrieval *quality* and clean, maintainable architecture — this is a real product, not a prototype. ## Required tech stack - **App:** Next.js (App Router), TypeScript, Vercel. Strong SSR/SSG, SEO, and JSON-LD structured-data experience. - **AI/backend:** Python; production RAG (embeddings, chunking, retrieval quality); LLM API integration. - **Data:** Postgres + **pgvector** (via Supabase); embeddings via a hosted model (Voyage/OpenAI). - **Auth & gating:** Supabase Auth with row-level security. - **Payments:** Stripe (subscriptions + one-time). ## Required skills - Shipped production Next.js (App Router) + TypeScript apps with strong SSR/SEO. - Built a real RAG / vector-search system in production — not a tutorial clone. - Comfortable in Python for data pipelines. - Postgres + pgvector and Supabase in production. - Stripe integration. - Plans before building; communicates clearly in writing. ## Nice to have - Audio/video transcription experience (Whisper / faster-whisper / Deepgram / AssemblyAI). - Agentic coding workflows (e.g., Claude Code). - Content-heavy SEO products or media libraries. ## Engagement - Hourly, contract-to-hire. ~20–40 hrs/week to start; long-term for the right person. - We start finalists on a **small paid test project** (a single self-contained slice of the pipeline) before the full engagement — that's how we evaluate fit. ## Confidentiality This is a proprietary product. Shortlisted candidates sign a mutual NDA before we share full scope and context. Please don't expect complete product details in the first exchange — strong technical applicants will have everything they need to be evaluated, and the rest follows the NDA. ## How to apply Applications that skip these are ignored: 1. **Start your proposal with the word `pgvector`** so we know you read this. 2. Link **two** projects: one live Next.js/SSR app, and one RAG/embeddings or LLM-integration project. Tell us what *you* personally built. 3. Answer briefly: *An offline embedding pipeline and a live search query must use the same embedding model — why does that matter, and how would you guarantee it?* 4. One line on your approach to chunking long-form audio/video transcripts for good retrieval.