Civic data · IndiaSoloConcept · building from Dec 2026
infoyantra
Indian government notices, made machine-readable: every rule sourced to its page, every change tracked.
The problem
infoyantra will collect official notices — recruitment notifications and their corrigenda first, then schemes and tenders — and turn them into structured data in which every field points to the page and clause it came from. It will watch them for changes. People will use it through change alerts and an eligibility check; software and AI assistants through an API and an MCP server.
Government information in India is public, but it arrives as PDFs — often scanned, often in Marathi or Hindi — spread across hundreds of sites. Working out whether you can apply for something means reading forty pages for the five lines that decide it.
Notices change after they are published. A recruitment notice can be amended twice before it closes, each time by another PDF on the same site. Nobody tells the people who were reading the first one.
Aggregator sites copy the headlines without the source, and official portals each cover one kind of document. Nothing joins them, keeps their history, or answers a question with the clause that decides it.
How it will work
- 1
Collect
Small, polite collectors per official site, run on a schedule. Every file is archived once, by its hash, with where and when it was fetched.
- 2
Read
Dates, fees, age limits, vacancies and eligibility rules are pulled out of each notice — rules first, a model only where rules can’t, in English, Marathi and Hindi.
- 3
Source
Every value carries the page, the clause and the words it came from. A value without a source is not shown.
- 4
Version
A corrigendum is linked to the notice it changes, and the change is worked out field by field: what moved, when, and on which page.
- 5
Answer
Change alerts, an eligibility check that cites the clause behind each answer, and an API and MCP server for software and AI assistants.
Decided up front
Sourced, not guessed
Every answer quotes the document it came from. When a case is borderline, it says “check the notification” instead of deciding.
Never official, never pretending
No government emblems, a clear “not affiliated” line on every page, and a link to the original document beside every fact.
Polite collection
Each site’s terms and rate limits are read and recorded before anything is collected from it. Login-walled or paid content is never collected.
One kind of document at a time
Recruitment notices and their corrigenda first, end to end, before schemes, and schemes before tenders.
The journey so far
What has happened, and the plan from here. Dashed dots haven’t happened yet.
Get annoyed
A notice that changed twice
A recruitment notice amended twice by corrigenda before it closed, each one another PDF on the same site. Keeping up meant checking by hand.
23 Oct 2026Research
The plan, written down
Who has the problem, what already exists, the first sources, the data model, the guardrails and the milestones — before any code.
Dec 2026BuildNext
Recruitment notices and their corrigenda
Collectors for about 25 official sites, sourced fields, version history, change alerts and an MCP server.
BuildNext
Schemes
The most-used Maharashtra and central schemes on the same engine, with an eligibility check that cites its clauses.
ShipNext
A tender pilot
Tender matching for a handful of small businesses, to learn whether the paid side works.
Building it
Planned stack
- Go
- PostgreSQL
- Cloud Run
- Cloudflare R2 + Workers
- Document AI
- MCP