DRISHTI.AI
Multimodal Deepfake Detection Platform
Six-layer forensic pipeline that detects deepfakes across image, video, audio and text — online or fully offline.
DRISHTI.AI is a multimodal deepfake detection platform built for the realities of Indian fact-checking: mixed media, patchy connectivity, and the need for explanations humans actually understand. It runs six detection layers in parallel, falls back to local mathematical forensics when cloud APIs are unavailable, and explains every verdict in plain English.
Single-modality deepfake detectors fail in the wild — manipulated content spans images, video, audio and text at once. Cloud-only detectors fail a different way: no connectivity, no verdict. Investigators need coverage, resilience and interpretability simultaneously.
- Six detection layers spanning image, video, audio and text, orchestrated as a LangGraph pipeline with parallel cloud API calls.
- Offline forensic fallback with zero external dependencies: DCT checkerboard analysis, Error Level Analysis, face-mesh asymmetry scoring, and STFT phase-discontinuity detection for audio.
- Neo4j graph persistence links scans by perceptual and forensic similarity, surfacing coordinated manipulation campaigns across submissions.
- Groq LLaMA 3.3 translates raw forensic signals into plain-English summaries an investigator can act on.
- Runner-Up, Cybersecurity domain — HackHive 2.0 “The Heist” (2026).
- Degrades gracefully: identical UX whether backed by cloud APIs or fully local math forensics.
- Every verdict ships with an explanation, not just a score.