The method behind the score.
How AI analysis works
FairLens runs a multi-stage pipeline on every application. The goal is to give reviewers a head start, not to replace human judgment. Each stage produces evidence the reviewer can audit.
1. Document parsing
Resumes, transcripts, and supporting documents are parsed into structured data: work history, education, skills, and accomplishments.
2. Text analysis
Essays and written responses are analysed for clarity, relevance, depth, and alignment with the programme's criteria.
3. Synthesis & scoring
Per-criterion scores are produced with reasoning and citations, then combined into a weighted overall score the reviewer can override.
Reading an AI score
Scores run on a 1–10 scale and are signals, not verdicts. Every score links to the source evidence in the application. Reviewers are encouraged to disagree. When they do, the override and reasoning are logged.
Fairness boundaries
Blind review support
Programmes can hide identifying information from reviewers during evaluation, reducing affinity bias.
Bias monitoring
Score distributions are tracked across cohorts. Drift is surfaced to organisers, not silently corrected.
Override audit trail
Every reviewer override is logged with timestamp and reasoning. Decisions stay defensible.
What we don't claim
FairLens does not eliminate bias, and no system can. The honest promise is structural: every score links to evidence, every override is logged, and the human reviewer holds the final call. That's the boundary, and we hold it.
Want a deeper walkthrough of the pipeline?