Capstone: ship a tiny AI app

You packaged the lane into a mini product: Python + data sense, a prediction or LLM call, maybe RAG or a tool — wrapped so it’s demoable. The artifact isn’t a folder of scraps; it’s a short script or slate path with a win condition you can name in one sentence.

Ship shape for a tiny AI app:

def run(user_input: str) -> str:
    # 1) prepare (clean / retrieve / score)
    context = user_input.strip()
    # 2) AI step (model, RAG, or tool loop)
    answer = f"Echo AI stub: {context}"
    # 3) return something a human can use
    return answer

print(run("Summarize: ship something small"))

1. One job — e.g. “answer from my FAQ” or “score this row”
2. One pipeline — input → prepare → AI → output
3. One demo script — happy path + what to say if it fails
4. One honesty check — limits, privacy, when to escalate to a human

Mental model: if you can’t demo it in two minutes, it isn’t shipped yet.

• Portfolio piece: “tiny RAG FAQ” or “tool-using helper”
• Internal prototype before a bigger build
• Teaching yourself by forcing a finish line

Scope creep kills shipping. Cut features until the happy path is solid. Never leave secrets in the repo. Label AI output as assistive when stakes are high.

Write a 15-second demo script: who the user is, what they click/type, what success looks like.