Prometrix
Companies use AI to filter candidates; Prometrix gives the candidate the AI. A full-stack product I founded that runs the whole job hunt, find, tailor, prepare, apply, with a human in the loop at every step.
The problem
Companies increasingly use AI to filter candidates, while the candidate is left with ChatGPT and a patchwork of paid tools. Prometrix puts the whole job search behind one AI: find the right roles, tailor the CV, write the cover letter, prep the interview, and apply. The aim is fairness: give every candidate the same kind of AI advantage large employers already have, used responsibly.
The non-negotiable principle: Prometrix generates, it never auto-submits an application and never fabricates experience. It reframes what is real, and a person approves before anything is sent.
What it does
Six stages of the job hunt, in one product:
| Stage | What it does |
|---|---|
| Find | Personalised job feed from your preferences (role, salary, location, visa sponsorship), broad UK coverage, and a daily match email. |
| Apply | One-click optimise-then-apply: evaluate the role, tailor the CV, draft the cover letter, then hand off to the real application. A browser extension does the same on any job board. |
| Score | A 10-dimension role-fit evaluation, a live ATS-style CV editor, and a free CV mini-audit. |
| Prepare | A mock-interview simulator with scored questions and a debrief, a LinkedIn profile optimiser, and a STAR story bank. |
| Track | An application tracker (applied to interview to offer) and a document vault for every CV and cover letter. |
| Autonomous ops | An agent-run company handles content and CV-service delivery overnight, built on an LLM agent SDK, behind hard quality gates. |
How it is built
A Python and FastAPI backend with async SQLAlchemy and background workers (for evaluation, alerts, and nightly agent runs), a Next.js, TypeScript, and Tailwind frontend, and a Chrome extension to optimise and apply on any listing.
AI runs across a tiered LLM setup: heavier models for evaluation and optimisation, lighter models for live scoring, with token and cost tracked on every call.
Trust by construction
The most important engineering is what the product refuses to do. Server-enforced quality gates block optimisation when the evaluator flags a fabrication risk or an unresolved critical gap, and nothing is ever auto-submitted or published.
That constraint is deliberate: the point is to strengthen a real application, not to invent one.
Status
Founded and in active development. The full flow has been verified locally end to end, from evaluation and optimisation through the quality gates to the apply hand-off. This repository is a public overview; the application source is kept private.