An AI onboarding assistant built at an internal AI hackathon, designed to answer new cybersecurity hires' procedural questions from an internal knowledge base instead of routing every one of them to a senior engineer.
Onboarding a new cybersecurity team member meant senior engineers answering the same procedural questions over and over — expensive for them, and slow for the person waiting on an answer.
A conversational onboarding assistant that handles the repeated procedural questions directly, grounded in the team's own documented procedures, with anything it can't answer falling through to a human.
An LLM front end over a set of domain-tuned system prompts, reading from the internal procedures knowledge base rather than from model memory — so answers track the documentation instead of drifting from it.
Designing an assistant that handles sensitive procedural knowledge responsibly — being deliberate about what it should answer confidently, and what it should hand to a person instead of guessing at.
Enterprise prompt engineering is mostly scoping, not wording: the useful work was deciding which questions the assistant owns and which ones it must escalate.
Source repository
Back to the portfolio