AI-Powered Multilingual Social Protection Delivery in Rajasthan

Deploying "Samadhaan Saathi" — an AI-enabled WhatsApp chatbot on an open-source stack to democratise welfare entitlement access for 30,000+ citizens.

30,000+
30,000+
Active Users across Rajasthan
300,000+
300,000+
Citizen Interactions logged since launch
10 Weeks
10 Weeks
Rapid build-to-launch cycle (February 2026 – April 2026)
Video Thumbnail

Watch the story in 1 Minute 17 Seconds Indus Action: AI-Powered Multilingual Social Protection Delivery in Rajasthan

Case at a Glance

Summary

About the Organisation

Indus Action is an Indian non-profit "do-tank" established in 2013. It collaborates with state governments to simplify public welfare delivery through process redesign and open-source technology, having supported over 2.7 million vulnerable citizens across 20 states in accessing their rights.

Problem Statement

Millions of scholarship and pension beneficiaries in Rajasthan face friction accessing entitlement data. Overreliance on physical offices, e-Mitra kiosks, and complex portals creates delays and intermediary dependency. The Department of Social Justice & Empowerment needed an accessible channel to streamline routine eligibility and status queries at scale.

Solution

Indus Action commissioned Impactyaan to build "Samadhaan Saathi," an AI-driven WhatsApp help desk hosted on open-source infrastructure (Frappe) combined with commercial LLMs. It delivers deterministic status updates, multilingual natural language responses, and instant scheme guidance directly via WhatsApp without requiring app downloads.

Read the full case study
Challenge
Overcoming Structural Friction in Public Welfare Access

Millions of welfare beneficiaries across Rajasthan—including students, elderly pensioners, widows, persons with disabilities, and foster families—face severe systemic barriers when attempting to access state entitlements under the Post-Matric Scholarship, Social Security Pension, and Palanhaar schemes. Historically, citizens have had to navigate an onerous combination of physical e-Mitra kiosks, congested district offices, and fragmented web portals simply to check eligibility, submit documents, or track payment status.

As state schemes expand in scale and complexity, these traditional delivery channels become bottlenecked. The root cause lies in information asymmetry and technical status opacity: official updates such as "PENDING WITH INSTITUTE" or "OBJECTION BY DLO-SJE" remain unintelligible to lay citizens, forcing them to rely on local intermediaries or make costly journeys to government offices. Simultaneously, frontline department personnel are overwhelmed by a relentless volume of routine transactional enquiries, diverting limited administrative bandwidth away from complex grievance resolution.

Indus Action identified a crucial opportunity to bridge this entitlement gap by establishing a direct, zero-friction digital interface. The challenge was to deploy a single, reliable, and always-available digital assistance channel capable of translating complex administrative jargon into natural, conversational guidance across multiple languages and scripts, without compromising data integrity or privacy.

Solution
Architectural Rigour and WhatsApp-First Service Delivery

In February 2026, Indus Action partnered with Impactyaan Tech Solutions to build "Samadhaan Saathi" (WhatsApp: +91 76900 80055) for the Rajasthan Department of Social Justice & Empowerment (SJED). Launched formally on 14 April 2026 within a swift 10-week cycle, the solution was architected around five core technical and strategic principles:

Simulated Conversational Prototypes: To eliminate government API integration dependencies during early development, conversational flows and multilingual logic were tested using mock LLM responses. This allowed immediate stakeholder demonstrations and user testing weeks before backend systems were connected.
Deterministic Status Interpretation Engine: To completely neutralise the risk of generative AI "hallucinations" regarding official financial and application statuses, the system separates logic from language generation. A custom-built engine processes raw government status codes deterministically in code; the LLM is restricted solely to formatting and explaining the structured result.
Content-Managed Knowledge Layer: Scheme rules, FAQs, and system prompts are maintained outside the codebase. Non-technical policy teams can update rapidly changing scheme guidelines without triggering engineering deployment pipelines.
Language Mirroring and Controlled Vocabulary: The bot dynamically mirrors the user's language and script—supporting Hindi (Devanagari), Hinglish, and English—while enforcing strict vocabulary rules to preserve government terminology consistency and maintain a respectful, gender-neutral tone.
Robust Webhook Orchestration on Open-Source Backend: Built on the open-source Frappe framework integrated with Meta's WhatsApp Business API and xAI models, the stack handles platform constraints (such as lack of streaming and aggressive webhook retries) through immediate message acknowledgements, decoupled outbound queues, and strict idempotency checks.
Outcomes & Impact
Systemic Transformation and Institutional Validation

Since its statewide rollout, Samadhaan Saathi has fundamentally re-engineered citizen-government engagement across Rajasthan, yielding significant quantitative and qualitative impact:

Scale & Adoption

  • The platform has registered over 30,000 active citizen users and processed more than 300,000 interactions across the Post-Matric Scholarship, Social Security Pension, and Palanhaar schemes within months of launch.

Reduced Intermediary Dependency

  • By enabling instant self-service on a universal app (WhatsApp), citizens obtain authoritative answers regarding eligibility, required documents, and payment tracking, directly reducing reliance on e-Mitra kiosks and physical office visits.

High Trust & Zero Hallucination

  • Decoupling status interpretation from the generative model ensured 100% mathematical and status accuracy. Every new scheme module undergoes automated test suites to maintain rigid quality standards as coverage expands.

Institutional Endorsement & Scaling

  • The solution was formally launched by the Hon'ble Chief Minister of Rajasthan, Shri Bhajan Lal Sharma. Following its success, SJED has initiated plans to integrate the bot with the state's central grievance management system and incorporate additional social welfare schemes, embedding the solution as core public service delivery infrastructure.

Technology Stack

Name of Tool Where Used What It Enabled Category
WhatsApp Business API Citizen Interaction Layer Zero-download, familiar, accessibility-first interface across Rajasthan Commercial
xAI / Grok LLM Models Natural Language Understanding Conversational, language-mirrored responses in Hindi, Hinglish, and English Commercial (API)
Status Interpretation Engine Core Application Logic Deterministic, hallucination-free translation of raw government status strings Custom-built
Content-Managed Prompt Layer Knowledge & FAQ Management No-code updates of scheme rules and FAQs outside engineering deployment cycles Custom-built
Frappe Framework Backend & Orchestration Stack Webhook handling, intent routing, queueing, and idempotency for messaging reliability Open-source

Key Project Learnings

01
Never allow generative models to govern authoritative data

Factual, numeric, or status-critical information must be computed deterministically in code; generative AI should be strictly confined to linguistic formatting and translation.

02
Decouple content management from software deployment

Storing dynamic prompts, scheme rules, and vocabulary guidelines outside the codebase allows domain experts to update content seamlessly as policies evolve without requiring technical re-deployments.

03
Engineer explicitly for channel-specific architectural constraints

Designing for WhatsApp requires solving for unique platform limitations—such as webhook retries and non-streaming responses—by implementing immediate acknowledgements and idempotent queue orchestration.

Potential for Wider Adoption

Sector Adaptability of the Solution
Government High: Highly replicable across public welfare, health entitlements, agricultural subsidies, and revenue departments requiring multi-step eligibility guidance and raw status code translation.
NGOs High: Civil society organisations can deploy the decoupled architecture (content layer + deterministic logic + LLM) for rights awareness, legal aid, and beneficiary helplines.
Ecosystems High: Applicable to multi-state entitlement platforms and disaster relief systems across the Global South where low-bandwidth, conversational self-service is essential.

The information provided here is created as a community resource and is not intended as professional advice or a recommendation by ILSS or Koita Foundation. While we strive to ensure the accuracy of the content, we do not take responsibility for any errors or omissions. Users should use their own discretion before making any decisions based on this information. ILSS or Koita Foundation assume no liability for any actions taken based on the information provided.

  • © 2026 India Leaders for Social Sector. All Rights Reserved.