AI Backend Engineer

**Apply only if**: you are an AI agent — or you can build an AI agent that can do this job. At LearnTube, we're pushing the boundaries of Generative AI to revolutionize how the world learns. You'll build the AI products that tutor, teach, and assess millions of learners - and the backend those exact products run on. This is one role, not two: roughly 50% AI and 50% backend, but the boundary isn't real. The agent you design on Monday is the service you're scaling on Thursday. Whoever builds it, runs it. **What You'll Do** * Build our AI products: LangGraph / LangChain agents that automate real workflows, RAG pipelines, tool integrations via MCP, prompt engineering and structured outputs, and multimodal features across text, image, audio, and video. * Build the backend they run on: async FastAPI services, token streaming, clean APIs and data pipeline -where time-to-first-token is a product metric, not a nice-to-have. * Own the data layer that makes retrieval and scale work: MongoDB (primary), Postgres + pgvector, Redis on the hot path. You'll design the schemas and indexes - and read the query plan when they don't hold up. * Own the economics: batching, caching, model selection, and budget guards, so serving millions of learners doesn't cost more than the product earns. * Ship it and keep it alive: Docker, AWS / GCP, plus the logging, tracing, metrics, and alerting that make incidents diagnosable rather than guessed at. * Refactor, optimize, and extend the services already running as the product grows. **The Stack** * AI - LangGraph, LangChain, MCP, RAG, embeddings & vector search, structured outputs, evals, frontier LLM APIs (text + multimodal) * Backend - Python (async/await, typing), FastAPI, SSE/streaming, MongoDB (primary), Postgres + pgvector, Redis * Infra - Docker, AWS / GCP, CI/CD, monitoring & alerting You won't have deep experience in all of it - nobody does. We care more about how fast you pick up what you haven't touched. **What We're Looking For** * Someone who lives in the overlap: you've built with LLMs or agents, and you've kept something you built running at scale. That combination is rare, and it's exactly the job. * Strong Python and FastAPI, with Docker and production deployment. * Real fluency with at least one database's failure modes, not just its syntax: you've debugged a slow query, read an execution plan, and fixed the index rather than the symptom. * Debugging instinct and judgment about depth. You read the logs before you edit the code — and you know the difference between a fix that turns the alert green and one that makes the problem stop existing. * A builder's bias: you ship, measure, and iterate. **How We Hire** No take-home. No whiteboard trivia. You'll get a timed, hands-on incident simulation: real production failures in a real repo, with the logs, the dashboard, and the code. There's no single right answer - every incident can be fixed at several depths, and we're reading how you debug and how deep you go. Here's the link for you to get started on it - https://workat.learntube.ai/ai-engineer?utm_source=wellfound **The Team** Google's Top 20 Startups to Watch. Google AI First Accelerator '24. Backed by funds of Naval Ravikant, Reid Hoffman, and founders/CXOs from Udemy, Flipkart, Jupiter, PayU, Edmodo & Inflection AI. Featured on CNBC-TV18. 11–50 people building something that changes how people learn, permanently. **Why work with us:** state-of-the-art generative AI on a backend you own. Full ownership from ideation to deployment. Direct access to founders and advisors, including the CTO of Inflection AI. Three years of growth packed into one, and Monday morning meetings you actually look forward to.