For judges
Two minutes, everything mapped to the criteria
Thank you for your time. This page maps each judging criterion from the rules to a claim and a proof you can open. It also says plainly what is real, what is sample data, and what is planned.
If you only check one thing, check this
Every score and figure on this site was produced by a script that is in the repository, and it writes down failures as readily as passes. One command, npm run audit, re-runs accessibility, Lighthouse, the tests, the build and every page, against this live deployment. The Proof page lists each claim next to the command that would catch us if it were wrong.
- 1Open Try it. Sunita's three papers and the are already loaded.
- 2Find tab: see the asset map, press Search here on one row to see the guided portal steps.
- 3Claim tab: pick SBI, answer three questions, press Get my route. Read the verdict and open the full .
- 4Download the claim pack PDF. Then open Track and Vault.
- 5Switch to dark mode, Hindi, and Technical mode from the top bar. Press any info button.
Criteria map
Where each criterion is proven
| Criterion (weight) | Our claim | See the proof |
|---|---|---|
| Innovation and Originality (25%) | The (AIS) asset map and the photo-to-claim flow exist in no other product. checks every route in the open. Institutions can bring their own rules. | Find tab: use the sample AIS; Claim tab: see the debate |
| Technical Implementation (25%) | with strict JSON schemas, a versioned with 5 institutions, an table parser, PDF claim packs, browser voice in and out, a public API with keys, and problem-details errors, light and dark themes. | API docs and playground; Technical mode on the home page |
| Real-World Impact (20%) | ₹1.84 is . Gujarat's camps returned ₹38,700 per family on average. One family like Sunita's keeps up to ₹63,000 that an agent would take. | Impact section and savings calculator |
| Feasibility and Scalability (15%) | One family costs about ₹36 of AI, measured against the live API rather than estimated, against an average of ₹38,700 returned per family at Gujarat's camps. No private data access is needed. Adding an institution is adding a file, not writing code. Families never pay; institutions, service centres and state camps do, and each of those has a stated reason to. | Further down this page, and the unit economics on the Proof page |
| User Experience and Design (10%) | Measured, not claimed: zero serious or critical violations across 11 pages in both themes, and on the live deployment across three runs of each profile: 99 to 100 performance on desktop and never below 90 on a throttled phone, with 100 accessibility, 100 best practices and 100 SEO on both, and layout shift between 0 and 0.024 against a 0.1 threshold. Full Hindi including the claim route and checklist, voice in and out, 44 px controls, info buttons everywhere, works at 320 px. | Test results on the Transparency page |
| Presentation and Demonstration (5%) | Sunita's story runs end to end in 3 minutes on sample data with one click. A reset button restores it. | Try it |
Why this is needed
What already exists, and what it still does not do
The government portals are good and we send people to them by name. They answer a different question: they let you search one place, if you already know what you are looking for. Nothing joins the picture together, and nothing helps with the part families actually get stuck on, which is the claim itself.
| Can a family do this today? | Government portals | A private agent | Virasat |
|---|---|---|---|
| Search banks, insurance, shares, dividends and mutual funds from one page | Yes, since 29 May 2026. The Common Landing Portal put one search front door over the separate portals, and it is a real improvement | Sometimes, for a share of the money | Yes, and this is not where we claim to be different |
| Turn a search result into a filed claim | No. Every portal searches and then hands the family back to the institution. The Supreme Court is hearing a plea about exactly this gap for legal heirs | Yes, for 10 to 30 percent | Yes. Route, documents, filled forms, tracking |
| Finds accounts the family never knew existed | No. You must already know the bank and the name to search | Rarely | Yes. The lists every institution that paid them |
| Reads a photograph of an old passbook or policy | No | No | Yes |
| Tells you which claim route applies, and why | No | Yes, but you cannot check the reasoning | Yes. A versioned , with the rule id shown |
| Fills the claim forms for you | No | Yes | Yes. A claim pack PDF, ready to submit |
| Works in Hindi, including the claim route and checklist | Partly | In person only | Yes, and it reads aloud |
| Tracks the claim and escalates if nobody replies | No | Sometimes | Yes. Escalation to the after 30 days |
| Cost to the family | Free | 10 to 30 percent of the money | Free |
Honesty
What is real, what is sample, what is planned
Real and working
- Document reading with AI (when an API key is set, or with your own key)
- PDF parsing by code
- for SBI, LIC, , and India Post
- Claim pack PDF generation
- (live with a key)
- Voice in and out in the browser
- Guided portal steps and deep links
- Public API with keys, and validation
- Light, dark, Simple and Technical modes
- Full Hindi: every screen, the output, and the
- Zero serious violations on 11 pages, live; 99 to 100 desktop and 90 or better on a throttled phone, with 100 accessibility, best practices and SEO on both
Sample data
- API are shared across instances only when Upstash Redis is configured; every response says which counter it used in X-Virasat-RateLimit-Store
- Sunita and Ramesh, their papers and amounts are made up
- The PDF is synthetic
- The unpaid-dividend index has made-up names in the same shape as real company lists
- Without an API key, the AI steps replay pre-computed results for the sample papers
Planned next
- WhatsApp channel
- login and certificate fetch
- consent for live bank data
- Marathi, Tamil, Telugu and Bengali
- A server database with OTP login for families who want to save across devices
- Real dividend lists fetched from company investor pages
In this prototype, all family data lives in the browser. Nothing personal reaches our servers unless the AI is asked to read a photo, and then the photo is not stored.
Deliverables
Everything the rules ask for
- Working prototype: Complete.This site. Try it with sample data in one click. Open
- Project description: Complete.The home page, sections Problem to Roadmap. Also a one page version in the repository. Open
- Source code: Complete.Public GitHub repository, MIT licensed, with the full commit history. Open (opens in a new tab)
- Documentation: Complete.README, technical design, UX spec and API docs. Open
- Demonstration video: Complete.Three minutes, recorded against this deployment in 4K. The script is generated from the file the narrator reads, so the two cannot disagree. Open (opens in a new tab)
- Pitch deck: Complete.The final deck, as a PDF you can open right now. Open
- Technology stack: Complete.Technology section, and Technical mode on every page. Open
- Demo link: Complete.This deployment, at virasat-indol.vercel.app. Open
- Audit report: Complete.The last full audit run, written down by the script rather than by us, including anything that did not pass. Open (opens in a new tab)
Feasibility
One family costs about ₹36.44 of AI. The family pays nothing.
A family never pays Virasat. A private agent typically takes a cut of whatever is recovered, which is exactly the tax on not knowing that this project exists to remove. So somebody else has to, and that answer has to survive a hard look. The figure above was measured against the live API with a cold cache, not estimated, and the working is on the Proof page.
Banks and insurers
A fee per completed claim pack that arrives correct the first time.
An institution already pays branch staff to handle these claims, and most of that time goes on packs that come back incomplete. A pack that is right on arrival is cheaper for them than the one they get today, and regulators already push them to bring balances down.
Not yet signed. This is the model we would take to a pilot, not an agreement we hold.
Common Service Centres
Virasat listed as one more service the operator can offer, earning the usual per service fee.
More than five of these centres already sit in villages doing exactly this kind of paperwork for a small fee. They are how rural India actually reaches a government service. We do not need families to find an app; we need operators to have one more thing on their screen.
The obvious distribution path, and the one we would test first.
State governments and the IEPF
Tooling for the camps they already run.
Gujarat's camps returned money to families at scale using volunteers and paper. The demand is proven and the budget already exists. What is missing is the software the volunteer sits in front of.
Closest fit to work already happening.
What has to be true for this to work
| It depends on | Where that usually goes wrong | What we did about it |
|---|---|---|
| No access to private or government systems | Most ideas in this space die waiting for an integration nobody will grant a student team. | Virasat needs none. It reads papers the family already has and public portals anyone can open. Everything is submitted by the family, through the official channel, exactly as it is today. |
| Handling personal documents lawfully | A service touching identity documents and account numbers is squarely inside the Act. | is the design, not a policy page: identifiers are masked at the moment of reading, family data stays in the browser, nothing is used for training, and one button deletes everything. Consent is asked for each step and is written in the language the family chose. |
| Adding institutions without rewriting the product | If each new bank needs engineering, this stops at five institutions forever. | A is a versioned file, validated against a published schema and served over the API. Adding an institution is adding data. Anyone can write one, and the endpoint will tell them whether it is valid. |
| Being wrong safely | A confidently wrong legal route could cost a family months in the wrong court. | The AI never picks the route; a tested does. The Referee can only add risks or ask for a person to look, never lower a caution. Anything outside the routes to needs-review instead of a guess. |
Scalability
Adding a new bank or insurer is just one new rule file
Start
1 district
Begin with service centres and one bank. Count the money returned every week.
State
All institutions
Add every bank, insurer and fund house in the state through .
India
22 languages
Connect and . Join RBI and IRDAI campaigns.
World
Beyond India
NRI families, and countries with similar programs.
Future scope
Where this goes next
Next three months
One district, measured in rupees returned
- Run it with operators and one bank branch in a single district.
- Publish one number every week: money actually reaching families, not sign ups.
- Watch where operators get stuck, because that is where the product is wrong.
Six to twelve months
Every institution in the state, by data not code
- Open the for contribution, with the schema and the validator already public.
- WhatsApp, because that is where families already are.
- More Indian languages, starting with the ones the pilot district actually speaks.
- An escalation path when an institution does not respond, since that is where families give up.
Beyond
The corpus becomes public infrastructure
- Consent-based discovery through and the , so a family stops having to remember what they owned.
- The claim rules adopted by whoever serves the family best, including the portals that already exist. We would rather be the standard than the site.
- Other countries run property programmes with the same shape. The corpus travels; the app is the easy part.