Where AI fits in this product

There is exactly one place we use generative AI: an optional feature that helps you draft the innovation statement required for DPIIT (Department for Promotion of Industry and Internal Trade) recognition. Nothing else on this site uses an AI model. Eligibility decisions, scheme ordering, translations, search — all of those run on rules, not on a model.

How the drafter works

You write your problem, your solution, your technology, and your scale. We send those four fields, plus a rubric that mirrors the official Inter-Ministerial Board scoring criteria, to Anthropic's Claude API. Claude returns a draft. You then read it, edit it, and decide whether to use it. We never submit the draft on your behalf. We never paste it into the official portal for you. Every output carries a visible "AI-generated draft. Always review before submitting." notice.

The model is instructed to refuse claims it cannot ground in what you provided — market sizes, competitive comparisons, fabricated technology assertions. If a section needs information you did not supply, the model leaves a placeholder rather than invent one.

What we send and what we keep

We send only the four input fields, the rubric, and the conversation history. No data from the rest of your profile is included in the prompt. No identifying details are added by the system. The full request and response are written to an audit log retained for seven years.

You can use this feature five times a day. The cap exists to keep cost predictable and to keep abuse contained.

Why this is the only AI surface

Generative models hallucinate. For most things this product does — telling you whether you are eligible for a scheme, sequencing your roadmap, citing a source — being wrong is worse than being slow. Rules and human-verified content beat a model on those tasks. The drafter is the one exception: a place where producing a starting draft for you to edit genuinely beats a blank page, and where the cost of a wrong word is low because you are the one who decides whether to use it.

We benchmark Indian providers regularly

Every six months we run a standing benchmark against Indian-region providers — Sarvam, AI4Bharat, Bhashini, Krutrim — on the same prompts and the same rubric. The day any one of them clears our quality, latency, and operational thresholds, we migrate to it.

See also

• Data residency
• Audit log
• Security and privacy

This summary is reviewed quarterly. Last reviewed: 28 May 2026.