BuildMe
3–5 hoursIntermediate

Build an AI That Actually Knows Your Stuff

Maps to: AI Application Builder · Knowledge Worker, Researcher, Solutions Engineer, AI Engineer

You're going to build an AI that answers questions using only your own notes, links, and documents, and that you can trust because every answer shows you the source it came from. The skill is grounding: getting an AI to answer reliably from a specific body of knowledge and verifying it against the sources instead of taking its word. That's RAG, one of the most in-demand things AI application builders do (every company wants an AI that knows their own stuff), and doing one tells you fast whether building trustworthy knowledge tools is your kind of work.

The plan

0/4 done

You're 20% in just for starting, the hardest part. Mark your first step done to keep the momentum.

  1. Decide what your agent should be an expert on: a topic you have real notes or links about. Load a first batch of sources and ask it one question. Seeing it answer FROM your sources, with citations, is the hook.

    Objective: A chosen corpus + a first answer with citations.

    1. 1

      Pick the knowledge area (a class, a hobby, a research topic) where you have real notes/links/docs.

    2. 2

      Load a first batch into NotebookLM and ask one question. It answers ONLY from your sources, with inline citations.

      Tool: NotebookLM

    Your call

    Choose what your agent should be an expert on (the corpus), yourself.

    What your agent should be an expert on, and why.

    What good looks like: Your agent answered a real question using only your sources, and you can click the citation to see where it came from.

    • Start with sources you know well. You'll instantly spot when an answer is off.

The bar to look back against

An AI agent that answers from your own corpus, and you can show with the citations which answers are genuinely grounded versus made up, having fixed what's in and out of the knowledge base when retrieval returned the wrong thing. The grounding is the work: not 'it answers questions,' but 'I can tell when it's grounded, and I fixed it when it wasn't.'

Finish the final step, then submit what you built. Your progress is saved.

Tools you'll use

Step 1 · Pick what it should know + load it + ask one question

NotebookLM Free

Google's tool that answers ONLY from the sources you give it, with inline citations to verify.

Best for: The free no-card RAG-lite default: load sources, ask, check citations for grounding. (Free: 100 notebooks, 50 sources each, 50 chats/day.)

Steps 2–3 · Build out the knowledge base

Dify Free

Open-source platform to build a real RAG agent (built-in retrieval/chunking).

Best for: The build-a-real-RAG route to go deeper (free cloud sandbox).

Claude or ChatGPT (Projects) Free

Upload docs to a Project and ask over them.

Best for: A simpler alternative: a Project with your docs attached.

How this shows up on a resume or college app

I built a RAG-lite AI agent that answers questions from my own notes and documents, and, by checking citations, learned to tell genuinely grounded answers from confident hallucinations, fixing the knowledge base when retrieval failed. I learned that the hard part of building AI on real sources is verifying it's actually grounded, not just that it sounds right.