Setu Intelligence Platform

Advanced Conversational AI & Analytics for the Karnataka State Police

Overview

Setu is an enterprise-grade, bilingual (Kannada & English) conversational AI designed exclusively for law enforcement. It empowers investigators to query vast crime databases using natural language, providing source-cited intelligence, advanced network visualization, and proactive pattern recognition.

Built natively on Zoho Catalyst, the platform ensures all intelligence is strictly grounded in case evidence.

---

Core Capabilities

• Bilingual Conversational Interface: Native support for English and Kannada via both text and voice.
• Explainable RAG: Retrieval-Augmented Generation provides deterministic answers backed by explicitly cited, clickable case sources.
• Criminal Network Visualization: Interactive D3.js node graphs identifying relationships between suspects, locations, and modus operandi.
• Tamper-Evident Auditing: Immutable hash-chained query logs ensure complete transparency.
• Strict Role-Based Access Control: Enforces jurisdiction boundaries automatically (e.g., Station Officer vs. State Analyst).
• High-Performance Hybrid Search: Proprietary District-Level Index Partitioning achieving sub-100ms retrieval latency at scale.

---

System Architecture

The architecture utilizes a serverless event-driven design, leveraging Zoho Catalyst Advanced I/O functions for orchestration, hybrid semantic search, and LLM synthesis.

graph TD
    classDef client fill:#000000,stroke:#333333,stroke-width:1px,color:#ffffff
    classDef serverless fill:#1e293b,stroke:#334155,stroke-width:1px,color:#ffffff
    classDef data fill:#0f172a,stroke:#334155,stroke-width:1px,color:#ffffff
    classDef external fill:#171717,stroke:#404040,stroke-width:1px,color:#ffffff

    User((Investigator))
    
    subgraph Frontend Client
        Chat[Web Application<br/>React + Vite]:::client
    end
    
    subgraph Serverless Backend
        API[API Gateway]:::serverless
        RBAC[Auth & Security Gate]:::serverless
        RAG[Hybrid Search Engine]:::serverless
    end
    
    subgraph Data Layer
        Index[(Local Partitioned<br>TF-IDF Index)]:::data
        Audit[(Catalyst<br>Audit Store)]:::data
    end
    
    subgraph AI Services
        LLM[QuickML LLM]:::external
    end
    
    User -->|Voice / Text| Frontend
    Chat -->|HTTPS Request| API
    
    API --> RBAC
    RBAC -->|Authorized| RAG
    RBAC -->|Log Request| Audit
    
    RAG -->|Filter & Score| Index
    RAG -->|Context Injection| LLM
    LLM -->|Synthesized Analysis| API
    
    API -->|Encrypted Payload| Chat

Loading

---

User Flow

sequenceDiagram
    autonumber
    actor Investigator
    participant UI as Setu Web Client
    participant API as Security API
    participant Index as Retrieval Engine
    participant LLM as AI Synthesizer
    
    Investigator->>UI: Voice/Text Query Input
    UI->>API: Authenticated POST Request
    
    API->>API: Enforce Jurisdiction (RBAC)
    API-->>Index: Scoped Request
    
    Index->>Index: Semantic & Structured Match
    Index-->>API: Extracted Case Sources
    
    API->>API: Mask Restricted Cases
    
    API->>LLM: Provide Sanitized Context
    LLM-->>API: Intelligence Summary
    
    API->>UI: Source-Cited Answer payload
    UI->>Investigator: Actionable Dashboard UI

Loading

---

Technical Stack

Component	Technology	Purpose
Frontend	React, TypeScript, Vite, D3.js	High-performance, reactive user interface and data visualization.
Backend	Python 3.10+	Orchestration, text processing, and security middleware.
Infrastructure	Zoho Catalyst	Serverless Advanced I/O, Data Store, and Edge Caching.
AI/ML	QuickML, TF-IDF, BM25	Hybrid retrieval and generative summarization.

---

Installation & Deployment

1. Repository Setup

git clone https://github.com/Monolithic-Dev/Datathon-Hack.git
cd Datathon-Hack

2. Catalyst CLI Initialization

npm install -g zcatalyst-cli
catalyst login
catalyst init

3. Dependency Management

# Initialize client environment
cd client && npm install

# Initialize serverless functions
cd ../functions/setu_api && pip install -r requirements.txt --break-system-packages

4. Local Development

# Boot the Catalyst local server
catalyst serve

# Start the frontend dev server (in a separate terminal)
cd client && npm run dev

---

Responsible AI Commitment

Setu is engineered with strict ethical guardrails. Predictive models and pattern recognition engines operate exclusively on modus operandi and case-level evidence. The system actively strips and prohibits filtering by demographic, religious, caste, or socio-economic indicators at the schema level.

Developed for the Karnataka State Police Datathon 2026