Evidence before scoring.
Instead of hiding everything behind a single “fit score,” the product uses a requirement-to-evidence map. Important role requirements are classified as demonstrated, partially demonstrated, or not evidenced, with source text visible to the user.
That distinction matters: “not visible on this resume” is not the same as “the candidate does not have this skill.”
From context to action.
The user adds a resume and a specific job description. The system extracts role signals, maps evidence, summarizes strengths and gaps, then lets the user select a gap and work through suggested edits or evidence prompts.
The goal is not an opaque verdict. It is a short, defensible list of actions the person can review, edit, accept, or reject.
AI reads. The candidate decides.
AI handles parsing, recurring requirement detection, evidence mapping, phrasing suggestions, and confidence estimates. The person decides what is true, relevant, worth emphasizing, and whether any suggested change still sounds like them.
Suggested rewrites must stay grounded in user-provided evidence. When evidence is missing, the system should ask rather than fabricate.
Useful, explainable, honest.
The prototype is successful if users reach a useful insight quickly, can trace recommendations back to source text, and leave with at least one concrete change they understand and agree with.
The core product constraint is integrity: no fabricated claims, metrics, skills, or experience.