2026-10-31

Bhoomitra — Voice-First Multimodal AI for Land Records & Civic Guidance

A voice-first civic AI assistant empowering rural citizens and farmers in Bihar to understand complex land records, mutation orders, and statutory survey notices in Hindi, Bhojpuri, Magahi, and English.

Inspect Source Code

The Problem

Land records, mutation orders, and special survey notices in Bihar are filled with complex bureaucratic legal jargon spread across dozens of circulars. Rural citizens and farming families often struggle to read official Hindi or English, creating heavy dependency on middlemen and missed statutory deadlines.

The Solution

Bhoomitra pairs Google Gemini multimodal vision with a deterministic statutory rule engine. Citizens speak or upload documents in their mother tongue (Hindi, Bhojpuri, Magahi, or English) to receive plain-language explanations, exact countdown deadlines, document checklists, and audio read-aloud.

Key Capabilities

Multilingual Voice Q&A: Speak naturally in Hindi, Bhojpuri, Magahi, or English with real-time audio soundwave animation and synchronized read-aloud.
Explain My Paper: Vision-based document inspection extracting case numbers, document types, required supporting papers, and circle camp counter numbers.
Zero-Hallucination Fact Grounding: Deadlines and land record searches are calculated strictly by deterministic code, while Gemini explains the findings in simple words.
Prapatra-2 Self-Declaration & Vanshavali Guide: Step-by-step guidance for family inheritance claims and statutory fee exemption alerts.
183 Passing Pytest Suites: Comprehensive automated test coverage for Pydantic request validation, OCR fallbacks, and dialect prompts.

Technology Stack

Python 3.11FastAPIGoogle Gemini AIReact 18ViteTailwind CSSWeb Speech APIPydanticPytest

Zero-Hallucination Civic AI Architecture

Bhoomitra divides responsibilities between deterministic code and multimodal LLMs:

[Citizen: Voice / Photo / Text]
             │
             ▼
      [FastAPI Router]
       ├── Code: Calculate statutory deadlines & query land records
       ├── Gemini AI: Multimodal OCR & plain-language dialect explanation
       └── Validation: Pydantic schemas & audit verification
             │
             ▼
[Plain Answer + Document Checklist + Next Steps + Audio Playback]

Key Technical Pillars

Multilingual Speech & Audio Follow: Browser-native speech recognition for Hindi/English with dialect audio playback and synchronous word highlighting.
Explain My Paper: Drops or photographs of official revenue notices trigger Gemini vision to identify deadlines (e.g., 21-day mutation response) and required evidence.
Rigorous Test Coverage: 183 automated unit and integration tests verifying audit logs, dialect prompts, and statutory rule compliance.

Lessons Learned & Key Takeaways

Building for rural citizens taught me that voice is the primary UI for public accessibility. Separating facts (handled deterministically in Python) from explanation (handled by LLMs) is crucial to eliminate hallucinations in civic and legal software.