Platform and integrations · In-house decision model
A small AI model of your own, not a chatbot
FitRank doesn't send your people to a chatbot to be ranked. A small model, trained for exactly these five questions, runs inside the product on an ordinary CPU.
The problem
Ranking people with a hosted chatbot costs money per call, returns free text you can't test, and can change whenever the provider updates it.
How it works
- 1An open-weight LLM labelled thousands of task–person pairs, and a small transformer learned its probability for every answer.
- 2The model answers five fixed questions per person and can't answer outside the options.
- 3Each company gets its own final scorer, calibration and cut-offs, learned from its own managers' decisions.
- 4It ships as Docker images for the API, worker and web app, with a Postgres-only job queue. No extra infrastructure.
What's included
An in-house decision model
A small transformer model, trained from an open-weight LLM, answers the matching questions. It runs on an ordinary CPU, so scoring costs nothing per run.
Calibrated per company
Each company gets its own final scorer, calibration and cut-offs, learned from its own managers' decisions.
Runs in containers
Docker images for the API, worker and web app, with a Postgres-only job queue. No extra infrastructure to run.
Where teams use it
- For leadership, hr and complianceMake staffing decisions you can audit and defendKeep a full trail of every recommendation, test recommendation rates for fairness, and never use protected attributes.
- For engineering and platform teamsAdd typed AI decisions to your own toolsAsk bounded questions over an API and get a probability for every option, with keys, budgets and rate limits per project.
Questions about in-house decision model
Does matching call ChatGPT or another hosted LLM?
No. Matching uses FitRank's own model on CPU. A hosted LLM is used only to read free text, such as a task description or a resume, into fixed fields.
How accurate is it?
On a synthetic benchmark of 40 unseen tasks and 647 people it put the best person in the top five 94.3% of the time, against 91.4% for rules only. Every pilot measures it on your own managers' decisions.
More of FitRank
- Decision APIYour own services ask FitRank typed questions over an API and get a probability for every option, with per-project keys, rate limits and budgets.
- Plain-language task intakeType the role you need in plain words. FitRank turns it into skills, level, dates and location, and you confirm every field before matching runs.
- Explained matching and shortlistsOne click ranks everyone who can take a task. Each person shows the rules they passed, the facts, the model's probabilities and the band they landed in.
Be one of the first teams to staff with FitRank
We're taking on a small number of pilot companies. Pre-register for early access, or talk to us about an annual licence.