IP-SAKTI Sahayak a multilingual, RAG-based (source-cited) AI assistant for Intellectual Property and regulatory guidance in Ayurveda, across national and international regimes.
Ministry of Ayush
Good pick. Genuinely. Now sit down, because the judges are going to try anyway — and this is what they will try.
Strong pick. A substantial, well-specified problem with a fully public corpus hiding under a Toys and Games label, and the formulation classification flow is what will distinguish you — build that and the citation grounding first, and let the international layer and multilingual delivery come only if time allows. Roughly 70–170 teams are expected to go here.
The receipts
Every red flag on this statement, in full. These are the four places it bites.
Exhibit A
Legal answers that are confidently wrong cause real harm, so uncalibrated generation is not merely a quality problem here — build the abstention path before you build the coverage
It gets worse
Retrieval assistants are a very common submission shape, so a build that answers fluently without visible citations and jurisdiction separation will be indistinguishable from many others
Still reading?
The corpus spans over a dozen national regimes and eight international instruments and the law moved recently with the 2023 biodiversity amendment, the 2024 patent and biodiversity rules and the WIPO GRATK Treaty, so version tracking with effective dates is a requirement rather than a refinement
And the finisher
The statement is enormous in scope; take its own advice and stage the build — a citation-grounded retrieval core over the Indian regimes done properly beats a shallow pass at both jurisdictions plus a knowledge graph
The damage report
Every score this statement earned, and what each one actually costs you.
Feasibility
4/5Actually buildable, which on this slate is rarer than it sounds. Do not squander it on scope.
Every source the corpus needs is genuinely public — India Code carries the statutes, IP India publishes the patent, trade mark, design and GI registries, WIPO and CBD treaty texts are open and TKDL has a public interface — and retrieval with span-level citation grounding is mature, so this is corpus curation and careful product design rather than a research problem.
Innovation scope
4/5There is something genuinely new here. Do not bury it under another dashboard.
The statement describes behaviours rather than mechanisms, and the genuinely original parts — a classification flow that determines the regulatory and IP posture before any question is answered, a jurisdiction toggle that refuses to conflate regimes, and calibrated abstention — are product design decisions nobody has made before in this domain.
Clarity
5/5The ask is unambiguous, which quietly removes your favourite excuse.
Remarkably thorough: it names every national and international regime in scope, sets out the formulation classification taxonomy with a worked contrast between a classical formulation and a new drug, specifies a staged build order, and states the evaluation criteria as answer accuracy, citation correctness, safe abstention and multilingual quality.
Acceptance potential
4/5Strong footing before you have written a line. Try not to waste it.
A strong hidden gem — the entire corpus is public and linked, the domain is specific enough that a generic retrieval build will visibly fail, safe abstention is written into the evaluation criteria so honesty is rewarded rather than penalised, and it sits under a Toys and Games theme label where no one will ever find it.
Effort
MassiveA semester of work wearing a hackathon costume. Something is getting cut; decide what now, not in week five.
Curating a version-tracked corpus across a dozen national regimes and eight international instruments, building the classification flow, grounding every answer to a citable span, calibrating abstention, adding multilingual delivery and meeting data protection and audit requirements is a large and unusually exacting build.
Demo-ability
EasyEasy to demo — and so is everyone else's. Working is the floor here, not the achievement.
A question in plain English returning a cited, jurisdiction-separated answer that a judge can click through to the actual statute is immediately impressive, and the classification flow asking the right clarifying question first demonstrates real domain reasoning rather than retrieval.
The demo they will have already seen
Somewhere around 70–170 teams are heading here, and the description is doing the choosing for most of them. They will read the same brief, reach the same architecture, and build a version of the same demo you are planning. Being correct is the floor. If your five minutes could be swapped with the team before you and nobody in the room would notice, you have not picked badly — you have built predictably, which costs exactly the same and hurts more.
What survives
The ground worth standing on when the questions start.
- The entire corpus is public and the statement links the sources, so unlike most legal AI ideas the data foundation is real and assembleable on day one
- The Toys and Games theme label is about as misleading as it gets, which makes a genuinely substantial statement invisible to every team browsing legal tech, AI or AYUSH
- Safe abstention is a stated evaluation criterion, so an assistant that correctly refuses an out-of-scope question scores rather than loses — this is unusual and it lets you build for honesty instead of coverage
Nothing here is fatal. It is just the list of places this statement pushes back, and you now get to push there first.
The framing is a joke. The findings are not — they are the same analysis on the statement page, and every line above is attached to a score or a fact in the record. It is one opinion with its reasoning attached, so argue with it before you trust it.