Skip to content
SSM
← All work

ARCHIVE · ENTRY 01 · CLASS: PUBLIC

SIH 2025 Winner2025

TattvaDrishti

AI threat-intelligence platform

An end-to-end platform that detects, attributes, and explains LLM-powered misinformation for analysts.

how it resolves
Content
Detection
Attribution
Risk score
Analyst dashboard
Problem

LLM-generated misinformation and coordinated influence campaigns are fast, cheap, and hard to detect, attribute, or explain. Analysts need more than a yes/no classifier — they need provenance, risk, and an audit trail they can defend.

Built

A full pipeline: fine-tuned DeBERTa-v3-Large models for AI-generated content detection and model-family attribution, layered with graph intelligence for coordinated-campaign discovery, explainable risk scoring, real-time SSE dashboards, and tamper-proof, blockchain-backed audit trails.

Role

Full-stack and system design — model integration, the analyst-facing interface, real-time data flow, and the audit layer.

Proof
  • Won Smart India Hackathon 2025 under the AICTE Cybersecurity Cell.
  • 95% accuracy on AI-generated content detection.
  • 98% accuracy on model-family attribution.
  • Graph intelligence + explainable risk scoring + tamper-proof audit trails.
Stack
FastAPINext.jsPyTorchDeBERTa-v3Graph intelligenceSSE
Field note · classified