Bias tests built on Indian CVs and interviews for AI hiring tools
TL;DR Evenlist sells bias test suites for AI tools that screen CVs or score interviews, built from Indian candidate material: regional English, caste and gender signals, college tiers. HR-tech vendors pay to show buyers abroad a clean audit, and large Indian employers pay before they switch a tool on.
AI screening is fast and its failures are quiet. A 2024 audit of text-embedding models used to rank CVs found White-associated names preferred in 85.1% of tests and female-associated names in only 11.1%[1]. Nobody has published the same thing for Indian names at that scale.
And the Indian bias may not even sit in names. A 2025 study scored 100 Indian and 100 UK interview transcripts with language models, and Indian candidates scored lower even with identities removed, because of sentence structure and vocabulary. Swapping names by gender, caste and region made no significant difference[2].
So a vendor that only runs name-swap checks would pass a tool that still marks down how Indians write and speak. That's a problem.
Evenlist keeps a library of paired test cases - the same CV or interview answer written in several Indian English styles, with signals for gender, region, caste-linked surnames, career gaps and tier-2 colleges changed one at a time. A vendor or employer runs its tool against the suite through an API and gets a report showing which signals shift scores and by how much.
The report follows the format US and EU buyers already ask for, so a vendor can hand it over during procurement. Employers get the same test before rollout and again every quarter.
The seed promised bias fixes across all of AI. I narrowed it to hiring because that's where buyers already face audit duties and where Indian test data is missing. Evenlist measures and explains. Fixing the model stays with the vendor.
There are two buyers. Indian HR-tech vendors selling into New York or the EU need a yearly bias audit or high-risk paperwork[3][4]. Large Indian employers, especially those using vendor tools for campus hiring, want evidence before a complaint lands. Nobody publishes prices for comparable audits, so my assumption is ₹4–12 L per tool per year.
New York has enforced bias audits for automated hiring tools since 5 July 2023[3]. The EU AI Act's high-risk obligations, which cover hiring tools, apply from 2 December 2027[4]. Indian HR-tech firms selling abroad need audit evidence before those deadlines, and language-model screening has spread fast since 2023.
Nobody publishes a figure for spending on hiring-AI audits in India. The narrowest signal is legal - every automated hiring tool used on New York candidates needs an audit within the year before use, with results posted publicly[3]. Next door, one AI-governance platform advertises more than 40 bias, safety and robustness tests as part of a governance platform[5].
AI governance platforms test for bias across many model types, mapped to the EU AI Act and ISO 42001[5]. Free open-source fairness toolkits are what teams use to measure and reduce unfair outcomes[6]. US audit firms run the New York audits. None that I found ships test data written in Indian English with Indian identity signals, and that dataset is the moat.
| Difficulty | Moderate (41/100) |
| Starting capital | ₹45–85 L |
| Time to launch | 6–9 months |
The software is modest. The hard part is building paired test material that linguists and sociologists accept as realistic, keeping it out of vendors' training data and earning enough trust that a buyer abroad accepts your report. Refreshing hidden test sets every quarter adds steady work too.
No licence is needed to test hiring tools for bias. The rules that matter are your buyers' audit duties abroad.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
Don't let AI decide on its own what counts as unfair. People with hiring and legal knowledge set the thresholds, and synthetic CVs can carry their own biases.
Want to work out the AI layer for Evenlist? Talk to Yash →
Here's what can go south -
I like this for a founder from NLP research or HR analytics with ties to Indian HR-tech vendors. If you're hoping Indian law will force demand soon then it's a poor fit. First prove that three vendors selling abroad will pay for a report and that a foreign buyer accepted one.
Plan for roughly ₹45–85 L, much of it building the test library, statistical work and selling to HR tech vendors.
No licence in India. New York's bias audit law and the EU AI Act create the demand, so check whether auditors must be independent.
Yes, and the US and EU are where buyers face audit duties today. Your edge is test data for Indian candidates.
The free Evenlist validation kit has the line-by-line capital breakdown, the effort scoring, the rules and data checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for vendors and employers, and 5 assumptions to test. Download the kit →
Thinking about building Evenlist or something close to it? The make-or-break work is a test library that experts trust and vendors cannot quietly train on, and I'm happy to think that through with you. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.