GovTech · Inclusion v1 uses lightweight, on-server AI

Welfare access is broken.
SARAL makes it fairer.

SARAL lets people apply by SMS/IVR on any phone, shows clear status updates, and runs simple fairness checks. The AI in v1 only assists (routing messages, suggesting likely schemes, spotting missing details); it never gives the final verdict.

Try SARAL Sandbox View Research Archieve

Multi-channel

SMS/IVR/USSD on basic phones

Transparent

Every step logged to your tracking ID

AI Assist

Routes intents; suggests missing info

Fairness checks

Simple patterns flag odd delays/denials

Note: v1 uses small models for intent routing and eligibility hints. Suggestions are non-binding and can be edited by staff. Personal data is minimised and encrypted at rest; operator actions are logged with reason codes. View model card.

SARAL high-level flow: citizen ↔ SMS/IVR ↔ operator dashboard

Current System

Opaque status; in-person follow-ups
App-first; basic-phone users excluded
High discretion → inconsistent outcomes

With SARAL

1-step SMS/IVR intake
Tracking ID + clear status page
AI hints + human review; all edits are logged

Model card (v1, brief)

Purpose: Helps route messages, suggest likely schemes, and spot missing details. It supports staff decisions, never replaces them.

Data & Privacy: Uses small, rule-based examples; avoids storing free text. Personal details are minimal and encrypted at rest.

Operator role: Staff can edit or override any AI hint. Every change is logged so reviews can see when and why something changed.

Reliability & Limits: Suggestions can miss edge cases or use simple language patterns. When unsure, the system falls back to fixed rules or asks for human review.

Improvement loop: Logged overrides are reviewed offline to refine rules and future versions.

System Failures

Barriers that stop citizens before they even reach the system.

Information Gaps

Eligibility rules and required documents are hard to find and often differ by office.

Digital Divide

App-only systems exclude citizens with basic phones or limited literacy.

Process Opacity

Citizens rarely know their application’s status or the reason for the delay; it simply “disappears.”

Discretion Risk

Unlogged steps and personal discretion lead to informal costs and inconsistent results.

Technical Solution

A bias-aware, SMS-first platform that turns every welfare application into a transparent, trackable journey.

SMS/IVR Intake

Works on any phone, citizens can apply or ask questions through simple messages or calls, in their own language.

AI Eligibility Assist

Small on-device models suggest likely schemes and flag missing information; staff can edit or confirm every suggestion.

Bias Monitoring

Simple analytics check for unusual patterns such as repeated delays or denials by location or group for human review.

Audit & Privacy

Every action by the system or operator is logged to a tracking ID. Personal data is minimised and securely encrypted at rest.

*v1 uses small, on-server AI models for routing and eligibility hints. Final decisions always involve a human check.*

System Architecture

Every message becomes a clear, traceable record — from intake to decision to delivery.

Normalize

SMS, IVR, or USSD inputs are cleaned and understood — the system detects language, extracts key details, and identifies intent.

Match

Rules connect citizens to the right welfare schemes and documents. AI hints appear only when confidence is high and always remain editable.

Monitor

The system checks for delays or rejections that look unusual. When patterns seem unfair, staff are alerted for review.

Notify

Each citizen receives a tracking ID and status updates with next steps. Messages clearly state whether an answer came from a rule or model.

  
Voices from the Ground
“I didn’t think it would work. Two months later — no more smoke.”Housekeeper, Ujjwala (Mumbai)
“The officer asked for money. Without paying, the housing papers wouldn’t move.”Villager, PMAY (Ratnagiri)
“I don’t have internet, but I think computers can be fairer than people.”Informal worker (Mumbai)

The Journey to SARAL

From SmartAgro (trust, mediation) → Project Nagrik (ground truth, context) → SARAL (structural response, transparency).

SmartAgro: Trust

Project Nagrik: Ground Evidence

SARAL: Auditable Access

SARAL Sandbox

A live demo showing how citizens interact with SARAL’s rule-based and AI-assisted workflows. Type a query like “ration card eligibility” or “housing scheme” to see how the system responds and how each decision is logged for transparency.

Terminal Demo (type and hit Enter)

SARAL: System ready. Try “ration card eligibility”, “pm kisan status”, “housing scheme”, or “ujjwala yojana”.

Audit Log

System Events & Fairness Checks

Waiting for events…

Help shape SARAL

SARAL is an early prototype exploring how citizens can get fairer access to welfare with less opacity, fewer barriers, and bias-aware checks. This is a long-term problem, and solving it at scale will take years, but every iteration is a step closer.

Note: This flow represents SARAL v1. In December, I’ll be conducting on-ground testing in India to stress-test the system with real users. Insights from that phase will guide future iterations.

Read the Blog Series What’s Project Nagrik? Share Feedback

Independent build · Informed by fieldwork & ongoing research · Open to feedback

Welfare access is broken.SARAL makes it fairer.