Watch the idea
The concept
Builder
RAG, agents, evals, APIs
Deployer
Sit with the customer and install
Product
AI PM · what not to automate
Domain + AI
A skill inside marketing, HR, ops
One artifact, three proofs
Proof of craft: it runs. Proof of judgment: it refuses when it should. Proof of taste: a human can review what leaves the building. Tutorials show you followed a video. A portfolio project shows you can be trusted.
Pick a painful, bounded job
HR policy assistant. Lab handbook bot. Support macros over real tickets. Not 'ChatGPT clone'. Bounded data, a user, a fail mode. That constraint is what makes the story interview-shaped.
Ship it where they can click
Live demo, repo, short video, and a one-page write-up: problem, approach, eval, what you would do next. If they cannot click it in five minutes, it does not exist in the hiring loop.
Worked example
The policy assistant
You take a public employee handbook, chunk it, retrieve, require citations, refuse off-policy questions, and publish a 30-case eval. That single project covers RAG, evals, and product judgment — the 2026 AI Engineer loop.
Picture to keep
Proof stack
- 1Role
- 2Artifact
- 3Metric
