AnAIcareer-coachingplatformwhereeveryinsightisgroundedinthemember'sownanswers.

Avoid rebuilding 'the user's intelligence' feature by feature; build the AI architecture once and reuse it across the platform.

Ensure outputs are personal and evidence-based — two members answering differently must see meaningfully different dashboards.

Deliver value inside the first five minutes for new members on a platform that was already live.

Give the client full cost control: capped hours per stage, written acceptance criteria, and the option to stop after any stage.

A single Career Intelligence Profile per member, with the resume, LinkedIn and interview tools plugged into it rather than rebuilt.

Thin AI layers — patterns, paths, strengths — generated only from the member's stored answers, with a rule that any insight unable to cite the member's own answers does not ship.

A personalised dashboard that shows what the platform has learned (snapshot, 'what we're noticing', next best move, career story) rather than only what the member has completed.

Product language kept to 'your responses suggest' — positioned as a preference exercise, not a psychological test.

React front-end with a Node.js API on Azure; one profile store feeding all AI features.

AI generation receives only the member's stored answers — no generic fallback — with safeguards against drift into generic output.

Each stage tests what it changed plus regression on affected journeys, on desktop and mobile; a full cross-browser end-to-end pass at the final stage.

Acceptance test: two test profiles answering differently must produce different strengths, paths and insights.

Baseline checklist recorded at stage one to protect both parties on regressions; regressions caused by new work fixed at no charge.