योजना ताई
Yojana Tai
A Marathi-first scheme copilot for Maharashtra, presented as a video agent, covering Ladki Bahin, PM-JAY with MJPJAY, MahaDBT scholarships and PM Surya Ghar.
- Languages
- Marathi · Hindi · English
- Channels
- Inbound and outbound phone · Missed call · WhatsApp · Browser video call
The problem
Maharashtra's welfare schemes are spread across departments and portals — MahaDBT, Aaple Sarkar, CSC counters, Anganwadis and Setu Kendras. A household that qualifies for four schemes usually hears about one, informally.
The drop-off then happens inside the application rather than before it: complex eligibility, a document nobody mentioned, a portal error, and a language barrier on top of all three.
Language is the part most often designed around rather than for. A Marathi speaker routed into a Hindi or English flow does not complain; they hang up, and the abandonment looks like disinterest in the data.
What AiSewak built
योजना ताई is a Marathi-first scheme copilot. She asks which language the citizen wants — Marathi, Hindi or English — and then holds it for the whole call, so the citizen never has to re-establish the language after every question.
The scheme coverage is specific to Maharashtra rather than a national list with a state label on it: Mukhyamantri Majhi Ladki Bahin, PM-JAY combined with the state's MJPJAY cover, PM Surya Ghar rooftop solar, PM-Kisan, MahaDBT farmer subsidies, MahaDBT pre-matric and post-matric scholarships, and PMAY-U alongside Ramai Awas.
Where the Uttar Pradesh sibling runs an audio orb, this page presents the agent as a video avatar. The same conversational agent drives both; the presentation differs because the audience and the device profile differ, and that is a deployment choice rather than a code fork.
The onward routing is named in the conversation itself — which paper, and whether the next stop is MahaDBT, Aaple Sarkar, a CSC, the Anganwadi or the Setu Kendra. The stated intent is no wrong queues, no middlemen and no fees.
How it runs
Conversational layer
An ElevenLabs Conversational AI agent on a voice profile selected for handling Marathi, Hindi and English in one conversation, rather than three single-language agents behind a menu.
Presentation
A video avatar widget shared with the Uttar Pradesh copilot, so a department can choose the avatar or the audio orb per channel without changing the agent.
Guardrails
The agent discloses that it is automated, never impersonates an officer, a minister or the Chief Minister, quotes only verified scheme facts, carries zero political content, and never asks for a full Aadhaar number, bank details or an OTP. Approval, amount and timeline are deferred to the department.
Compliance review
Automated compliance evaluation runs on the whole conversation set rather than a sample, with transcripts and structured outcomes retained for audit.
What a department can replicate
- 1
The published pilot shape is one or two districts — the demo names Mumbai Suburban and Nashik — carrying five to ten flagship schemes across voice and WhatsApp, with MahaDBT and Aaple Sarkar deep links, before any statewide commitment.
- 2
The scale-up assumes a standardised scheme knowledge-base template, a call-centre copilot for officers, and district dashboards for Collectors and Zilla Parishad CEOs.
- 3
The transferable decision here is the language one. Auto-switching from the caller's first words, and then locking, is cheaper to build than a language IVR and removes the step where most callers in a regional language leave.
Scope the same pilot for your department
Two districts, a bounded knowledge base, a written list of what the agent will refuse to do, and a measurement step in the call itself. That is the shape every pilot above started from.