All roadmaps
intermediate
~3 months · 5-8 hrs/week· 7 stages

LLM App Builder

The 'AI engineer' path — code an LLM app end to end

You'll have shipped an LLM-powered app with RAG and streaming, know where each piece fits, and speak the language fluently.

Who this is for

Comfortable with basic programming (Python or JS). You want to build real AI apps, not just use them.

Prerequisites
  • Write and run basic Python.
  • Comfortable with a terminal and git.
North-star project

Ship a 'chat with X' app (your notes, a PDF library, a codebase) — streaming, RAG-grounded, cited, deployed.

Your progress0 of 7 · 0%

Progress is saved on this device. Sign in to sync it across devices.

  1. 1

    Stage 1: Python you'll actually use

    The minimum viable Python for AI work — the rest you'll pick up on demand.

    Deliverable · Load a CSV with pandas, filter it, and print a summary in under 20 lines.
  2. 2

    Stage 2: Talk to a model from code

    Your first real AI code — this is the atomic unit of every LLM app you'll ever build.

    Deliverable · Ship a 50-line CLI that streams answers from an LLM about anything you paste in.
  3. 3

    Stage 3: Give it memory

    Every chat, every agent, every assistant needs conversation state. It's simpler than it sounds.

    Deliverable · Extend your CLI into a persistent chatbot with a running conversation history.
  4. 4

    Stage 4: Embeddings & semantic search

    The math behind RAG, recommendations, and 'similar to X' — used everywhere.

    Deliverable · Embed 50 short docs. Given a query, return the 5 most similar. No LLM yet.
  5. 5

    Stage 5: RAG: ground the model in your data

    The single most useful pattern in production AI. Everything else is a variation on this loop.

    Deliverable · Turn your chatbot into a RAG bot over a set of docs you actually care about.
  6. 6

    Stage 6: Reliability & evaluation

    The difference between a demo and a product. Measure quality, then improve it on purpose.

    Deliverable · Build a 20-question eval set for your RAG bot. Fix the top 3 failure modes.
  7. 7

    Stage 7: Ship it

    A thing that lives on the internet with your name on it is worth 100 side projects that don't.

    Deliverable · Deploy your RAG app publicly. Write a post about what you learned. Link it in your bio.