Source Open the bot

Code for a Billion 2026 · Livelihood

Know before you go.

A Hindi and English chat that tells an unorganised worker which welfare schemes fit, what to carry, and where to begin — before a day's wage is spent finding out at a counter.

Try it on Telegram See how it decides
Telegram · हिन्दी

आपके लिए 3 योजना मिलीं। सबसे ऊपर वह है जिसमें सबसे ज़्यादा फ़ायदा है।

प्रधानमंत्री श्रम योगी मानधन

क्या मिलेगा: 60 साल के बाद कम से कम ₹3,000 महीना पेंशन।

कहाँ से शुरू करें: नज़दीकी जन सेवा केंद्र (CSC)

यह रकम मिलने की गारंटी नहीं है। आवेदन और अधिकारी की जाँच बाकी है।

Two minutes — what the project is and how a screening runs

The problem

One wrong answer costs a day's wage.

Schemes are not hidden — they are published and announced. What a worker cannot tell from the outside is whether a scheme will actually accept them.

43.99 crpeople working in India's unorganised sector, 2019–201
31.48 cralready on e-Shram, with fourteen central schemes mapped to it2
₹455average daily earnings, male casual labourer, 20253
₹315average daily earnings, female casual labourer, 20253

Registration is not receipt of a benefit — none of these figures measures this project

How it decides

A verdict ledger, not a black box.

Each screening selects its scope first, then runs the same cited rules against the answers it needs.

01 — Input

Choose, then answer

Workers can screen all verified schemes or select only the ones they want. The flow asks only what that choice needs.

02 — Evidence

Cited rules

One readable TOML file per scheme, each condition beside its official source and review date.

03 — Decide

Pure rule engine

Same answers, same verdict. No network call in the path, no language-model import in the module.

04 — Explain

Next step

The reason behind the result, the documents to carry, and the kind of counter to start at.

Built to refuse guessing

  • A missing answer becomes UNKNOWN, never a default value.
  • An unsigned or unfinished rule becomes UNKNOWN, never a real verdict.
  • Insurance cover and annual pension are never summed into one flattering number.

Built to be checked

  • Rules are plain text files in the public repository, not hidden prompts.
  • Every value records its government source and the date a person read it.
  • Boundary tests compare the engine against source interpretations encoded separately, by hand.
The WhatsApp bot offering Hindi and English as tappable buttons, then stating it will not ask for a name, phone number or Aadhaar number, and that only counts are kept.
Fig. 1 — Buttons before typing, and the privacy promise made up front
A Telegram result listing two schemes, each with what you get, why you qualify, and where to go — including the premium and what the cover excludes.
Fig. 2 — Every verdict carries what you get, why, and what it excludes
The answer recap showing unanswered facts written as 'you did not say' and 'needs confirmation' rather than being filled in with a default.
Fig. 3 — “You did not say” survives all the way to the verdict

The design

UNKNOWN is an answer.

Most screeners have two outcomes and quietly default the third. If a worker has not answered a question, or a rule has not been checked by a person, this one says so — and names the missing fact instead of implying the worker failed a test.

ELIGIBLEINELIGIBLEUNKNOWN

Architecture

The whole system, on one screen.

The optional language model sits outside the decision path. It may propose an occupation label for the worker to confirm — and nothing else.

The complete Yojana Sathi core flow: a worker opens Telegram, WhatsApp or preview; selects language; gives consent; chooses all verified schemes or selected schemes; answers only the needed questions; builds a temporary profile; and receives deterministic eligible, ineligible or unknown verdicts from human-signed TOML rules. The flow then explains results, documents, an application pack and where to go. Privacy-safe metrics retain only consented coarse events and an optional self-reported pilot role.
Fig. 4 — The complete core logic: selected schemes reduce questions; the deterministic engine decides eligibility
The WhatsApp bot offering to make a sheet, then delivering it as a file, saying it has no name and no Aadhaar on it — only what to ask for and what to carry.
Fig. 8 — The sheet, offered and never forced

The handoff

What the worker walks away holding.

  • A sheet built in memory — no copy is kept on the server.
  • Papers they said they do not have, turned into questions to ask at the counter.
  • No name, no Aadhaar on it — only what to ask for and what to carry.
  • Optional feedback includes a self-reported pilot role, and says so: “you can skip it.”

Coverage

Ten signed schemes. Still narrow by design.

The Uttarakhand widow-pension rate remains in review. A reviewed set of answers that hold up beats a broad directory of approximate ones. National pension figures below are the stated central share; state pensions can be higher, and overlapping pension routes are never added together.

SchemeStated benefitWhere an application begins
PM Shram Yogi Maandhan₹36,000 per year from age 60Common Service Centre
Uttarakhand old-age pension₹18,000 per yearState portal
Uttarakhand widow pension₹18,000 per yearState portal
PM Jeevan Jyoti Bima Yojana₹2,00,000 life coverBank branch
PM Suraksha Bima Yojana₹2,00,000 accident coverBank branch
PM Ujjwala YojanaIn-kind LPG supportOnline or an LPG distributor
e-Shram registrationGateway — no payout of its owne-Shram portal or a CSC
Indira Gandhi National Old Age Pension₹2,400 per year central share; ₹6,000 from age 80Common Service Centre
Indira Gandhi National Widow Pension₹3,600 per year central share; ₹6,000 from age 80Common Service Centre
Indira Gandhi National Disability Pension₹3,600 per year central share; ₹6,000 from age 80Common Service Centre
National Pension Scheme for Traders₹36,000 per year from age 60Common Service Centre
Python stdlibTOML rulesSQLiteTelegram Bot APIWhatsApp Cloud APIDockerUbuntu + systemdOne small server

The honest section

What this page will not claim.

Naming what has not been shown yet is part of the product, not a disclaimer buried at the bottom.

No field impact yet

No worker has used this outside maintainer testing. No claim about applications filed, benefits received, or people reached.

Amounts are entitlements

Every rupee figure is what a scheme states it pays — not money anyone has been shown to have received through this tool.

No government endorsement

An independent open-source project. Where a rule here conflicts with a department, the department is right.

One rate needs confirming

The Uttarakhand widow-pension amount comes from myScheme because the department page does not state it.

Language needs field review

Both flows are tested by the maintainer. Comprehension has not been checked with non-technical workers.

WhatsApp is not general

Telegram is live. WhatsApp runs on Meta's test tier until business verification completes.

Sources

  1. Ministry of Labour & Employment, Lok Sabha reply, 24 July 2023: 43.99 crore workers in the unorganised sector.
  2. Ministry of Labour & Employment, 2 February 2026, on registrations as at 26 January 2026: 31.48 crore workers on e-Shram, fourteen central schemes integrated.
  3. National Statistical Office / MoSPI, PLFS Annual Report 2025 press note: daily casual-labour earnings.
  4. Scheme-by-scheme evidence, source snapshots, review records and tests: the public repository.