Career Projection Generator: Building an Evidence-Bounded AI Automated Workflow
[!NOTE] Classification Level:
PUBLIC - THOUGHT LEADERSHIP— Technical Architecture Series.
The Problem: Career Context Without Positioning Inflation
I wanted to build a system that could take my professional experience and turn it into a role-specific interview playbook but ask an LLM to do that and it will quietly round up.
AI tools are great at drafting role-tailored content — proposals, playbooks, CVs. "Adjacent experience" becomes "direct experience." "Familiar with" becomes "expert in." The model isn't lying, it's just optimistic by default.
For me, this isn't just a CV problem. It's the same challenge I run into building any AI application that draws on a trusted knowledge base: how do you let a model adapt facts to a new context without letting it drift from what's actually true?
Every interview needs its own angle, the same experience gets framed differently depending on the role. Doing that by hand, one prompt at a time, means redoing the same setup work every time, and each pass risks a slightly different (and slightly inflated) version of your story.
So I built the Career Projection Generator: a multi-agent pipeline that takes one locked set of career facts and reshapes them for each target role — with the honesty rules enforced in code, not just in the prompt.