AI Engineer Intern

About the role

We’re an early-stage edtech startup building an AI-powered adaptive learning product. We’re looking for someone to own the quality of our AI layer — building on an existing system to ship a new data-driven feature end-to-end, along with the evaluation harness that proves it works. Backend infra is handled; your job is to make the AI genuinely good.

Why it’s worth your time

Not fetch-the-coffee intern work. You’ll take a real feature from raw data to shipped, on a live product with real users — and push the AI layer’s quality as far as you can take it. If you’d rather build something in AI than watch tutorials about it, this is that.

You’ll come out having shipped a production agentic RAG system end-to-end — the hands-on experience hiring teams now screen for and that’s hard to get from coursework. Plus a strong reference, full credit for your work, and a real conversation about continuing if it goes well.

You don’t need to grind 24/7 — we care that what you build is good.

What you’ll do

• Build and structure the AI system’s knowledge bases — extract, clean, and organize information from multiple sources into high-quality, retrieval-ready form
• Build the data ingestion + tagging pipeline that feeds them
• Design and tune the retrieval engine — chunking, embeddings, vector DB, retrieval quality
• Wire up tool-calling and multi-agent orchestration across the workflow
• Ship a new data-driven metric on top of the existing system
• Build the evaluation harness — eval loops, regression checks, and handling the cases where the model gets it wrong
• Extend and harden the AI layer — prompt/context design and orchestration

Required

• Strong Python — clean, readable code; can read, debug, and extend an existing codebase, problem solving skills, familiarity with git operations.
• Solid conceptual understanding of how LLMs work — tokens, embeddings, context windows, why models hallucinate, model quality/evaluation(weights and metrics), VLMs/MLLMs.
• LLM APIs (OpenAI / Anthropic / Gemini) — prompt engineering, structured output, streaming, and function/tool calling, rate limiting.
• At least one real agentic AI project — multi-step or multi-agent orchestration, tool calling, or a self- correction loop. A real GitHub repo, not a tutorial clone.
• RAG depth — chunking, embeddings, retrieval quality, and how to debug bad retrieval
• Eval literacy — you can measure whether output is actually good, not just that it runs
• Experience structuring messy, multi-source data into clean, usable form
• Pursuing or recently completed a degree in CS, AI, Data Science, or a related technical field.

Bonus

• Agent frameworks — LangGraph / LlamaIndex / CrewAI / Autogen / OpenAI Agents SDK, or a hand-rolled agent loop
• Solid CS fundamentals; a write-up or paper about something you built
• MCP (Model Context Protocol), or other agent-integration / tool-connection frameworks
• Tier-1/2 CS program (IIT / IIIT / BITS / NIT or similar)

Details

Remote (India) · 2–3 months · ₹10,000–20,000/month (based on skills) · Strong LoR + feature ownership + portfolio-able work + possible continuation

How to apply

• Fill out this form: https://forms.gle/Ep192aFAFF8yYKFLA
• Link your GitHub, and tell us the most interesting AI project you’ve built — what it did, and how you handled it when it broke or behaved unexpectedly.
• We read every submission.