Two years ago a portfolio with a chatbot in it stood out. It no longer does. Here is what has replaced it.
Retrieval, evaluation, cost
Employers ask three questions about an LLM project: where does the context come from, how do you know the answers are right, and what does it cost per thousand requests? A project that answers all three beats a cleverer one that answers none.
- Retrieval: how documents are chunked, indexed and cited
- Evaluation: a test set, and numbers that moved
- Cost: tokens per request, and what you did to reduce them
Show the failures
Interviews turn on the question of what went wrong. Keep the examples where your system answered confidently and incorrectly, and be ready to explain what you changed.