Catalogue

Prompt Engineering & LLM

Designing, hardening and evaluating instructions for large language models.

Getting a correct first answer with a single instruction and an explicit output contract.

Decomposing hard tasks into legible steps and anchoring the final answer.

Probing a model for harm, documenting reproduction steps and rating severity.

Testing instruction hierarchies and refusal calibration against multi-turn pressure.

Building instruction datasets that survive dedupe, balance and leakage checks.

Judging retrieval-augmented answers on recall, groundedness and citation accuracy.