Human judgement, AI speed, evidence for everything
Our AI evaluation workflow keeps a human in control at every step that affects your score.
1. You submit your product
Share a product description, website or chatbot endpoint, API documentation, sample conversations and testing requirements — through the enquiry form or your customer portal.
2. Requirements analysis
A project manager confirms scope, languages, success criteria, timelines and the access you are authorising. Nothing is accessed without your explicit written permission.
3. AI-generated test scenarios
Our platform drafts realistic Indian scenarios — Hinglish typos, frustration, regional terms, multi-turn context, safety probes.
4. Human approval
A QA reviewer approves, edits or rejects every suggested scenario. Only approved scenarios can be executed.
5. Automated execution (where authorised)
If you connect an authorised chatbot or agent endpoint, approved cases can be run automatically and repeatedly. Automated results are never scored until a human reviews them.
6. Human evaluation
Verified testers matched by language, accent, region and device run the scenarios on real phones and record evidence: transcripts, screenshots, audio and notes.
7. Issue identification
Failures become tracked issues with severity, reproduction steps, expected and actual results.
8. Quality scoring
The Bharat AI Quality Score is calculated from reviewed evidence only, using documented category weights. Incomplete evaluations are marked provisional.
9. Report generation & approval
A branded report with executive summary, coverage, scores, failed cases, evidence and recommendations is reviewed and approved before delivery.
10. Client delivery
You receive the report in your portal, can drill into every test case and issue, and compare future releases.
Five minutes of your time
- A product link, app, phone number, WhatsApp number or chat link (or an API endpoint for automated runs)
- Who your customers are
- 2–3 tasks or journeys you want tested
- Test login details if needed — never real customer accounts
Typical human test: brief approved and paid on day 0, testers matched on day 0–1, sessions recorded on days 1–2, reviewed report on day 3.
How we keep results trustworthy
Real, verified people
Every tester is 18+, email-verified, assessed for their languages and profiled. No bots, no AI personas.
Recorded evidence
Results without evidence are rejected. Every finding links to a transcript, recording or screenshot.
Quality scores
We rate every submission. Testers with low scores stop receiving tests.
AI that saves time, not truth
AI suggests scenarios and drafts summaries from real transcripts. A person reviews every scenario, result, score and report.
No public reviews
Testers never post ratings or reviews, keeping you within consumer-protection rules.
Data protection
Consent before recording, limited access, encryption, and recordings deleted 90 days after delivery.
How testers take part
1. Join and verify
Apply, verify email and complete a short language assessment.
2. Accept tests
Receive offers that match your languages, city and phone; accept or decline.
3. Record and submit
Do the tasks, speak your thoughts, and submit your results with evidence.
4. Get paid weekly
Approved tests are paid every week by UPI.
Ready to see your score?
Most evaluations start within a few working days of access being confirmed.