RABI delivers real-time legal guidance and live government document access in spoken Hindi — an AI lawyer on the surface, a document intelligence layer underneath. RABI is optimised for the 1 Billion+ people in the world's most populous country for whom document access and quality legal advice was previously out of reach.
India's legal system functions in English, bills by the hour, and assumes procedural literacy that most of the country was never allowed to access. For most Indians, the qualified legal opinions they need to navigate life are priced, located, and worded beyond reach.
Filings, and records operate in a language alien to most litigants. Most cases are lost before they begin due to this discrepancy and the uneven access to information it generates.
A certified copy of a land record is sold for as much as 1000+ Rupees by middlemen. The market prices out the working-class Indian.
Rights, limitation periods, and procedure are ostensibly public information. In reality, they are distributed unevenly. People only learn what they are entitled to after the window to claim it has closed.
RABI is operated entirely by voice, in Hindi, over a phone call inside the app. Underneath the conversation runs a document intelligence layer that retrieves, reads, and produces the paperwork relevant to the conversation from government databases — live today on the Uttar Pradesh Bhulekh system, and extending soon across the Indian Government's API Setu catalogue.
Real-time, full-duplex legal consultation in spoken Hindi. The LLM conducts a structured intake, identifies the legal issues in the caller's own account, and explains rights, procedure, and next steps at the caller's pace, with support for interruption and clarification throughout.
Every consultation concludes with a Raasta — a written, shareable action plan summarizing the matter, pinpointing the applicable provisions, and explaining the concrete steps ahead, generated after the call from the full conversation transcript.
Live retrieval of government records through a brief structured call. RABI retrieves the caller's khatauni directly from the state portal during the conversation, resolves the caller's district, tehsil, and village by voice, and returns the document with a plain-language reading of what it contains and what in it needs attention.
The retrieval layer is being extended across the API Setu and DigiLocker catalogue — certificates, registrations, and entitlement records.
Structured drafting of complaints, applications, and notices from the consultation record. Output fields are verification-gated: the system does not emit a name, date, or amount it has not verified, and unverified fields are delivered as explicit blanks for the citizen or their advocate to complete.
RABI features an iterative frontend reduced to a single surface, and a highly-capable backend that executes the entire task while the caller is on the line.
The entire interface is one card that floats over the call. It writes down what RABI has understood in real-time, offers a tap when a choice has to be made, becomes a number-pad when a number is needed, turns into the search as it initiates, and conveys the successful location of the record. No menus, no forms, no second screen - a revolution in UI Design that optimizes for the particular end-user's actual usability patterns.
Built according to Nielsen's usability heuristics for a user who is local-language-first and button-first: visibility of what the system is doing, recognition over recall, one action at a time.
End to end, during the call: the caller's district, tehsil and village are resolved against an index of the state's own place tree, the holder is found by name or number, and the khatauni is pulled from the Uttar Pradesh Bhulekh system — typically in under a minute, with the caller confirming the result in one tap.
The extraction layer is built for a government portal that does not behave like a typical API: results are kept across the fleet, so a slow village or a dropped line never loses a record. The record then generates a खतौनी जांच report — every finding graded red, yellow or green.
A real-time voice intelligence stack on the frontend, an agentic document infrastructure on the backend.
Native Hindi speech-to-speech over a live, full-duplex session — sub-second response, barge-in handling, and session resumption across network interruptions. The conversation is the interface: the only thing ever typed is a plot or account number, on a dialer on the card, so it is never misheard.
Headless retrieval agents drive government portals in real time during the call, with structured extraction, disambiguation against portal records, and in-session document capture. Retrieved records are rendered, explained, and delivered as the खतौनी जांच report the moment the call ends.
The system operates under a strict grounding rule: it does not assert the contents of any record it has not retrieved in that session. Legal citations are constrained to verified sources, and document fields that cannot be verified are surfaced as blanks rather than guesses.
Served from Google Cloud with a native Android client, engineered to ride out portal throttling without losing a caller's search. Personal data handling is aligned with India's DPDP framework; conversation data is processed for service delivery and never sold.
RABI is currently operational across Uttar Pradesh — a state of more than 200 million residents, with the country's largest volume of land and revenue disputes. Live land-record retrieval is integrated with the state's Bhulekh systems across all 75 districts and 350 tehsils.
RABI expands state-wise as each integration reaches production reliability. Additional states are under evaluation.
Every figure below is drawn from RABI's own production records.
World-class legal guidance and official document access, delivered in the caller's own language, to over a billion people in India.
Aditya is a public-policy practitioner, trained at Columbia University (MPA), with field experience across district-level administrative and revenue systems in North India — the terrain RABI operates in.
He builds and operates RABI end to end: model orchestration, the document processing layer, state-portal integrations, and distribution. The product's direction is set by continuous analysis of its own production conversations, and by multi-stakeholder engagement spanning government data systems, the startup ecosystem, and the communities RABI serves.
The khatauni retrieval capability is a direct result of that analysis. A structured review of production transcripts showed land and revenue matters to be the single largest category of demand, with callers asking for the record itself alongside advice about it. That finding set the build priority, and live land-record retrieval shipped as a consequence — the product refining itself through what its users actually discuss.
RABI has been recognised by Youth Co:Lab, a UNDP and Citi Foundation initiative, the Atal Innovation Mission, GSF Founders Academy, and the INNOSTARS acceleration program at IISER Mohali, and is supported by Google for Startups with USD 2,000 in Google Cloud credits.
RABI is free to use and live on Google Play.
