Generally available
One prompt across every Act, Rule, state variant, policy and record
Regulens Copilot
- 94%
- of answers rated correct in blind expert review
- 0
- uncited factual claims permitted in a response
- 12 sec
- median time to a cited answer
Compliance questions do not respect module boundaries, or state boundaries. Neither should the way you ask them.
The problem
What this module exists to fix
Answers live in five systems
The rule is in one place, the obligation register in another, the SOP in SharePoint, the licence in a filing cabinet, and the data in your ERP. Answering one question means five searches and a reconciliation.
General AI assistants are confidently wrong about Indian law
A model without a grounded corpus will produce a fluent answer citing a section that does not exist, or reasoning imported from a foreign regime that does not apply here. In this domain that is worse than no answer.
Institutional knowledge is not queryable
The reasoning behind past interpretations — including how a specific regional office applies a rule — sits in email threads and the memory of people who have since left.
Capabilities
What it does
Grounded, cited answers
Every response is grounded in retrieved source material and carries citations to the specific provision, obligation, policy paragraph or record it relies on. Uncited assertions are suppressed.
Cross-module reasoning
A single question can traverse regulations, obligations, impact assessments, policies, controls, licences, BRDs and — where connected — transaction results.
Permission-aware retrieval
The copilot can only see what the asking user can see. Entity, site and document-level permissions are enforced at retrieval, not filtered after generation.
Refusal over invention
When the corpus does not support an answer — including where a state has not yet notified its rules under a Central Code — it says so and shows what it searched. Calibration is measured and published in your tenant quality report.
Saved queries and monitors
Turn a question into a scheduled monitor that alerts you when the answer changes.
Full interaction audit
Every prompt, retrieval set, response and citation is logged for review, quality assurance and model-risk governance.
How it works
Step by step
- 01
Interpret
The question is decomposed into retrieval intents across the relevant modules.
- 02
Retrieve
Hybrid semantic and keyword retrieval runs over the permitted corpus, scoped to the asking user and their entities and sites.
- 03
Ground
Candidate passages are ranked and checked for sufficiency. Insufficient grounding triggers a refusal, not a guess.
- 04
Answer
The response is composed with inline citations to each supporting source.
- 05
Verify
A verification pass checks each claim against its cited passage before the answer is shown.
Questions
The things people actually ask
Including the ones where the answer is a limitation.
- Do you train models on our data?
- No. Your content is never used to train shared or third-party models. Retrieval operates over your tenant-isolated index and inference runs under zero-retention terms with our model providers.
- How do you prevent hallucination?
- Three mechanisms: retrieval is restricted to a curated corpus, generation is constrained to cited passages, and a separate verification pass checks each claim against its citation before display. Where grounding is insufficient the copilot declines and tells you what it searched — including declining to answer with foreign-law reasoning where Indian law is silent or a state rule is not yet notified.
- Can we govern it under our model risk framework?
- Yes. We provide model documentation, evaluation results, drift monitoring and full interaction logs suitable for internal model risk governance.
See Regulens Copilot against your own footprint
A two-week scoped pilot on your real entities and jurisdictions. You compare the output against what your team found in the same period.