Citizen Services Portal with Multi-Language NLP

The Challenge

A state government was operating 14 separate citizen service portals — one per department — each with its own login system, form format, and status tracking mechanism. Citizens applying for a building permit, a business licence, and a vehicle registration had to create three separate accounts, fill out redundant personal information three times, and check three different websites for status updates.

The portals were only available in English, despite 62% of the state's population primarily using regional languages. Call centre volume was 40,000+ calls per month, with 70% of calls being simple status inquiries.

Our Approach

We built a unified citizen services portal that consolidates all 14 department workflows behind a single Next.js frontend with a common identity layer (Aadhaar-based eKYC for authentication).

Workflow Engine: Each department's service workflow was modelled as a finite state machine in a custom workflow engine. The engine manages form submission, document upload, department review, approval/rejection, and citizen notification. Workflows are defined in YAML and can be modified by department administrators without engineering involvement.

Multilingual NLP Chatbot: We deployed a chatbot supporting 8 regional languages (Hindi, Tamil, Telugu, Kannada, Malayalam, Bengali, Marathi, Gujarati) plus English. The chatbot handles form assistance (guided form filling through conversational Q&A), status queries ("Where is my building permit application?"), and FAQ resolution. Language detection is automatic, and the chatbot maintains context across language switches within a single conversation.

Notification System: Citizens receive SMS and WhatsApp notifications at each workflow state transition, in their preferred language. Push notifications are sent via the portal's PWA for users who have installed it.

Results