← All workCase study No. 12AI Assistant · Web + Mobile + Voice12 / 17

MunshiJee.

An AI legal assistant you can talk to

An AI legal assistant for India, available on the web, on mobile, over WhatsApp and by voice.

Client: MunshiJee
Industry: LegalTech / AI
Engagement: Build
Relationship: A client partnership

munshijee.in

MunshiJee 3D voice assistant in a library setting with a tap-to-talk microphone button

Visit munshijee.in

MunshiJee sign-in screen on a phone — Mobile web Sign-in on a phone, the way into the assistant.

IThe ground

The brief.

MunshiJee set out to make accurate, factual legal information accessible to people in India — in English and Hindi, and through whichever channel suits them: the web, a mobile app, WhatsApp or a voice call. Answers had to be grounded in real statutes and judgments, and the experience had to feel approachable for people who might never open a legal database.

IIThe plan

Our response.

We built a Next.js web app with a 3D, lip-synced avatar that people can simply talk to, powered by Vapi voice assistants in English and Hindi. Behind it, a NestJS backend runs an agentic chat with legal tools — web and scholarly search, statute lookup, judgment search and similar-judgment discovery — and reads uploaded documents with OCR. The same assistant answers on WhatsApp through Twilio, generates legal drafts and case reports as PDFs, and runs on OpenRouter with feature flags, budget caps and fallback providers. A Flutter app brings it to mobile.

AI Agent Development
Voice AI
3D Avatar Experience
Web Development
Mobile App Development
WhatsApp Integration

IIIThe buildStrategy → craft → engineering → care

Brick by brick.

Course 01
Talk, Don't Type
A React Three Fiber avatar with audio lip-sync and Vapi voice assistants in English and Hindi, so a legal question can simply be asked out loud.
Course 02
Grounded Answers
An agentic chat with streaming responses and a tool registry for legal search, statute lookup and judgment discovery, plus document upload with OCR.
Course 03
Every Channel
One assistant across the web, a Flutter app, WhatsApp through Twilio and voice calls, with legal drafts and case reports delivered as PDFs.
Course 04
Built to Run
OpenRouter orchestration with budget caps and fallbacks, Razorpay subscriptions, push notifications, an admin console and 52 backend test suites.

IVThe outcomeWhat took shape

What it became.

Channels: 4; Web, mobile, WhatsApp and voice
Voice Languages: 2; English and Hindi voice assistants
Agent Tools: 7; Legal and scholarly search, statutes, judgments, uploads and links
API Endpoints: 97; Across 16 NestJS controllers

Project scope and implementation details from Edifio’s case study.

VSpecificationBuilt on the right foundations