A data scientist who moved from building models to testing them.
Anything later stays hidden. The site also remembers the chapters you open on this device.
File AIPR-018Major · Herald
Role
Data scientist on the Hyderabad site's governance team
Work
Data scientist at NovaChem. She joined in early 2025 building models, moved to testing them, and joined the site's three-person governance team in December 2025.
Home
Grew up in a village outside Tirunelveli, Tamil Nadu.
Family
Father drove an auto-rickshaw. Mother taught at the government school.
To find where a system is unfair and to be sure of the finding.
Needs
To stay independent of the people who depend on the systems she tests.
Stands to lose
Applicants like her who are marked down unseen, and her own independence.
Strength
Rigour. She records how small her samples are, has others rerun her search, and says what she cannot prove.
Flaw
Her own history points her where to look, which she knows is a reason to look and no evidence.
Wound
She grew up outside Tirunelveli, the daughter of an auto-rickshaw driver and a government schoolteacher, and studied at a state university and by distance learning. NovaChem's old hiring screen turned her away twice before a person read her third application.
The question their story asksCan the person who was once wrongly screened out test the screen fairly?
Herald. She brings the finding that forces everyone who signed for the system to decide what they owe.
The arc
FromA tester running a monthly blind audit.
→
ToThe auditor who has found one harm, refused a move that would compromise her, and is already searching for the next.
In July 2026 she finds that the hiring system marks down applicants where three ordinary traits meet. Her own CV scores in the lower third. She argues against stopping the system, asks that recruiters read without the score, and declines a post in human resources so that she does not answer to the people who depend on the system.
part of the arc is hidden until you have read Chapter 31.
In July 2026 her search over combinations of traits finds a group the system scores far below what careful human readers give them. She checks it on a fresh blind sample and notes how few people each sample holds.
She runs her own CV through the test copy and it scores in the lower third. She tells Meera the system would have rejected her a third time, and still argues against stopping it.
The pull. A post in the department that owns the system, or her place outside it.
The choice. She declines the post.
The cost. The offer and whatever it would have brought.
She holds that an evaluator must not answer to those who depend on what she evaluates.
later choices are hidden. They appear as you read on.
What they learn
A system can be clean on each trait and unfair where traits meet. A personal reason to look is not evidence of what will be found. Chapter 31
hidden until you have read further.
In the text
“The person evaluating the system must not answer to the people who depend on it.”
later lines are hidden.
People
Meera Reddysenior manager to whom her analysis goes
Her analysis lands on Meera's desk as eleven printed pages.
They disagree openly about stopping the system and work together on the re-reading.
the site's head of human resourcesowner of the system she tested
She tells him what she cannot prove and refuses his offer of a post in his department.
What they carry
Revathi carries the book's point about independent evaluation and about harm at the overlap of ordinary traits. She is also one of the people a system judged and nobody went back to, which is the debt the chapter is about.
Part VIFound the applicants the hiring system marked down where three ordinary traits meet.
Chapter 31: The Moral Responsibility of BuildersRuns the monthly blind audit and, in July 2026, the search over combinations of traits. Checks the result against a fresh blind sample and records how few people each sample held. Runs her own CV through the test copy; it scores in the lower third. Argues against stopping the system. Asks that recruiters read without the score and that the team record how far down the order each reading went. Declines an offer to move into human resources. At the end her search is running again on a different set of traits.