Should Kestrel screen campus applicants using their social media?
Last hiring cycle, 18% of freshers who accepted an offer never joined, and 22% of those who did left within a year. Talent Acquisition wants a way to predict both before an offer is made. The proposal is to buy Orbis SocialFit, which reads a candidate's public social media and returns a Reliability score, a Culture-fit score and a Professionalism flag within minutes of application. HireLens Review is the agent that examines whether that information is suitable, representative and properly used for this purpose, and tells the decision owners what it found.
- Head of Talent AcquisitionWhether the business case justifies the tool, and whether culture fit should be scored at all.
- Data Protection OfficerWhether the processing is lawful and proportionate, and the conditions it must meet.
Autonomous review run
Press Run the review live and the agent works through the case on its own: it reads the case file, puts every signal through the six questions, opens candidate files, runs the fairness test, checks the law and research, and hands to named people anything outside its authority.
You can watch each step as it happens. It finishes with a memo for the two decision owners. A live run takes two to three minutes. Replay the recorded run plays back a real earlier run, step by step.
- 1Read the case file
- 2Review the signals
- 3Investigate candidates
- 4Run the fairness test
- 5Check the evidence
- 6Hand off to people
- 7Apply the decision rules
Do not adopt as proposed. Orbis SocialFit should not be used to score or shortlist campus applicants.
Five layers, from the model up
- Quality controlsAutomated tests on every rule and number, a documented test log, and a named human sign-off on the decision sheet.
- Reliable executionScores, outcomes, selection rates and the recommendation come from fixed rules in code. The agent calls them; it never invents them.
- Enterprise contextThe case file, the vendor's documents, and an approved library led by Indian law: the DPDP Act and Rules, the Code on Wages, and Puttaswamy.
- Domain methodSix questions every signal must answer, seven outcome rules, three decision rules, and explicit points where a person must decide.
- General model capabilityThe agent plans the review, reads the evidence, explains it in plain language, and drafts the memo.
Check the agent’s work
All thirteen signals through the six questions, with the rule that decided each.
Nine sample cases: wrong person, tagged posts, misread languages, absence penalties.
Selection rates by group across 1,000 synthetic applicants. Switch signals on and off.
Indian law first, then reference standards, research and precedent.
The recommendation, the gaps, the human decisions, and the sign-off.
What the agent may and may not do, and what this review cannot show.