ZynoviqInsights
What we learn, published carefully.
We publish what we learn about how AI handles Indian professional work — on public models, with our methods — so the whole field can improve. We never publish anything about a customer's system.
Planned publications
| Title | Type | Planned for |
|---|---|---|
| India GST AI Report, edition 1 | Report | March 2027 |
| Method note: how certified answers are made | Method | January 2027 |
| Method note: the confidence score | Method | January 2027 |
| Why AI judges aren't enough for professional work | Paper | April 2027 |
Research principles
- Public models only in published comparisons.
- Methods disclosed, with honest limits.
- Separate case sets from any customer's private tests.
- No league tables of customers. Ever.
- Corrections published if we find an error in our work.
Research topics we're working on
- How often AI is confidently wrong on Indian tax, and whether it signals doubt
- Which kinds of legal reasoning — exceptions, recent changes, facts — AI finds hardest
- How quickly AI answers go out of date after a rule change
- Measuring uncertainty handling on genuinely contested questions
Collaborate with us
We welcome research partnerships with universities and institutes on methods for evaluating AI in regulated professions. Contact us.
Our research — FAQ
Do you publish results on customers' AI systems?+
Never. Published research covers only publicly available AI models.
Can universities collaborate with you?+
Yes, on methods for evaluating AI in regulated professions.
What if you find an error in published research?+
We publish a correction.