Prajna-AI (ಪ್ರಜ್ಞಾ-AI)

Conversational Ai Copilot & Biometrics platform for the Karnataka State Police

Prajna-AI turns raw FIR records and case data into instant, conversational crime intelligence for officers — searchable in English, Kannada, or Hindi — backed by a retrieval-grounded LLM pipeline that never invents facts.

---

What it does

Instead of manually digging through case files, an officer can ask Prajna-AI a question in plain language:

"Show vehicle theft cases in Bengaluru" "What are the crime trends this year?" "Who are the known repeat offenders in this district?"

The system detects intent, extracts entities (district, crime type, status, FIR number, etc.), queries live FIR data in MongoDB, and hands the retrieved records to an LLM that is explicitly constrained to answer only from that data — no invented FIR numbers, statistics, or case details.

Key Features

Module	Description
Conversational Crime Intelligence	RAG-based chatbot grounded in real FIR data, multilingual (EN/KN/HI), voice input support
Dashboard	At-a-glance case load, active investigations, hotspots, and repeat-offender stats
Real-Time Crime Center (RTCC)	Live operational overview
Hotspot Intelligence	Geographic crime clustering on an interactive Karnataka map
Offender Profiling	Suspect/offender profiles built from linked FIRs
Network Analysis	Criminal network graph — associates, gangs, linked cases
Trends & Analytics	Monthly trends, crime-type breakdowns, district comparisons
Sociological Insights	Age/gender demographic analysis of accused and victims
Recidivism Engine	Repeat-offense risk scoring
Predictive Insights / Forecasting & Alerts	Forward-looking risk and hotspot forecasting
Financial Analysis	Mule-account/transaction network visualization
AI Face Search & Suspect Identification	Biometric face matching against a convict database
NAFIS Fingerprint Matcher	Fingerprint-based identification workflow
Operational Simulation / AI Tactical Commander	Scenario planning and tactical recommendations
Decision Support / Explainable AI	Reasoning transparency for AI-assisted decisions
Reports & Audit / Governance	Audit log tracking and compliance oversight
Citizen Services	Citizen incident reporting, DigiLocker verification, wanted-sighting reports
User Management & Role-Based Access	Investigator / Analyst / Supervisor roles

Architecture

┌─────────────────┐      ┌───────────────────────┐      ┌──────────────┐
│  React Frontend  │─────▶│  Catalyst Advanced I/O │─────▶│  MongoDB     │
│  (Vite + TS)     │ /server proxy│  Function (Node 18)   │      │  Atlas       │
└─────────────────┘      └───────────┬───────────┘      └──────────────┘
                                      │
                    ┌─────────────────┼─────────────────┐
                    ▼                 ▼                 ▼
             Intent Detector   Entity Extractor    Query/Filter Builder
                    │                 │                 │
                    └─────────────────┴─────────────────┘
                                      │
                                      ▼
                         Analytics + Context Builder
                                      │
                                      ▼
                    QuickML (GLM-4.7-Flash) — grounded LLM response
                                      │
                                      ▼
                              Frontend chat UI

A separate Python face-matching microservice (face-matching-service/) and a Zoho Zia speech function (zia_speech) handle biometric search and voice transcription respectively.

Tech Stack

Frontend: React 19, TypeScript, Vite, Tailwind CSS, React Router, Recharts, Leaflet (maps), jsPDF, html2canvas

Backend: Node.js 18, Zoho Catalyst (Advanced I/O serverless functions), MongoDB Atlas

AI / ML: Zoho QuickML (GLM) for conversational intelligence, custom face-matching service (Python)

Infra: Zoho Catalyst Cloud Scale (API Gateway, Authentication/Whitelisting, Data Store)

Project Structure

Prajna_AI/
├── Prajna_Frontend/           # React + Vite SPA
│   ├── src/
│   │   ├── components/        # chat, charts, maps, biometrics, network, etc.
│   │   ├── pages/              # one page per feature module
│   │   ├── services/api.ts     # backend API client
│   │   ├── utils/api.ts        # low-level fetch helper (proxy-aware)
│   │   └── context/             # Auth, Language, Toast providers
│   └── vite.config.ts          # dev-server proxy → Catalyst backend
│
├── Prajna_Backend/
│   ├── functions/
│   │   ├── prajna_ai_function/  # main RAG pipeline (Advanced I/O)
│   │   │   ├── index.js          # request entrypoint
│   │   │   ├── intentDetector.js
│   │   │   ├── entityExtractor.js
│   │   │   ├── executeQuery.js   # MongoDB filter + query
│   │   │   ├── analyticsEngine.js
│   │   │   ├── responseGenerator.js
│   │   │   ├── aiEngine.js       # QuickML integration
│   │   │   ├── offenderEngine.js / networkEngine.js / trendEngine.js / hotspotEngine.js / riskEngine.js / scenarioEngine.js
│   │   │   └── db.js             # MongoDB connection (+ in-memory fallback)
│   │   └── zia_speech/          # speech-to-text function
│   └── catalyst.json
│
└── face-matching-service/      # Python face recognition microservice

Getting Started

Prerequisites

• Node.js 18+
• A Zoho Catalyst project (Cloud Scale) with a MongoDB Atlas cluster
• Python 3.12 (only if running the face-matching service locally)

1. Frontend

cd Prajna_Frontend
npm install
cp .env.example .env   # fill in values, or leave VITE_API_BASE_URL empty to use the dev proxy
npm run dev

2. Backend (Catalyst function)

The backend runs as a Zoho Catalyst Advanced I/O function. To deploy:

cd Prajna_Backend
npm install -g zcatalyst-cli   # if not already installed
catalyst login
catalyst deploy

Configure these environment variables in the Catalyst Console (Settings → Environment Variables) for your project/environment:

Variable	Description
MONGODB_URI	MongoDB Atlas connection string
PROJECT_ID	Catalyst project ID
ORG_ID	Catalyst org ID
CLIENT_ID / CLIENT_SECRET / REFRESH_TOKEN	Zoho OAuth credentials for QuickML (from api-console.zoho.in)

Also confirm in API Gateway that the deployed function's API has Authentication: No Authentication and, if calling from localhost during development, that your dev origin is added under Authentication → Whitelisting → Authorized Domains with CORS enabled.

3. Face-matching microservice (optional, local dev)

cd Prajna_Backend/face-matching-service
pip install -r requirements.txt
python app.py

Environment Variables (Frontend)

Variable	Purpose
VITE_API_BASE_URL	Backend base URL. Leave empty for local dev (uses Vite's /server proxy); set to the deployed Catalyst URL for direct/production use.
VITE_SPEECH_API_BASE_URL	Speech-to-text function endpoint
VITE_FACE_MATCHING_URL	Face-matching microservice URL (defaults to local proxy at 127.0.0.1:8000)
VITE_ENABLE_MOCK_FALLBACK	Whether UI components fall back to mock data when the backend is unreachable

Data Grounding

All conversational responses are generated only from FIR records actually retrieved from MongoDB for the current query. If no records match, the system states that plainly rather than fabricating an answer — this is enforced in the LLM system prompt in aiEngine.js and the context passed by responseGenerator.js.

Multi-language Support

The chatbot and reports support English, Kannada (ಕನ್ನಡ), and Hindi (हिन्दी), including full report generation and UI translation.

Built For

Karnataka State Police (KSP) — developed as part of the KSP Datathon by Team BITHEADS.

---

This README reflects the current state of the codebase. For architecture decisions and setup troubleshooting, see inline comments in db.js, aiEngine.js, and vite.config.ts.