Prompting and Engineering with LLMs (Colab)

Research-track 1-of-N option, distinct from the Agents/Coding round: a Colab-based 55-minute exercise centered on designing prompts for a defined task and engineering the surrounding evaluation/scaffolding code. Requirements From recruiter blurbs: "A 55-minute coding on prompting and engineering with LLMs in Colab." "It is LLM-related, but we don't specifically test your LLM knowledge here. We'll give you all the context you need in the interview." Reported shape: Colab, Google Meet, screen-share. A task is described in the notebook; you design and iterate on prompts to solve it. Some variants have a debug component (a Colab cell that runs but produces wrong outputs). Engineering pieces: evaluation harness, prompt template management, handling structured outputs, possibly chaining calls. Notes This round is sometimes confused with the Agents/Coding round. The recruiter blurb is the disambiguator: Agents/Coding mentions "Tool Use and Agents," Prompting & Engineering does not. Open-book on docs; AI assistants typically not allowed. Specific task details have not yet been publicly documented; treat this card as a structural overview and re-check for fresh writeups close to your interview date. Preparation Practice writing a structured-output prompt (XML or JSON schema), parsing the response, and routing it to downstream code. Build a small evaluation loop in Colab: 20 inputs, run with two prompt variants, score automatically, surface diffs. Be comfortable using client.messages.create directly from a Colab cell — no SDK ceremony, no syntactic memorization required.