This is a submission for the MLH x DEV Writing Challenge
What I Built
Suryakavacham: Evidence Before Verdict.
What happens when a video, audio clip, or image cannot be trusted at face value?
I built Suryakavacham, a multimodal media integrity and forensic verification platform designed to help analysts investigate potentially manipulated or unverified media.
Instead of making a simple AI-generated real-or-fake judgment, Suryakavacham brings together independent signals from vision, audio, temporal patterns, metadata, and provenance to support evidence-based investigation.
Key features include:
- Multimodal analysis: Correlates independent forensic signals with explicit uncertainty.
- Privacy-focused evidence handling: Uses AES-256-GCM envelope encryption for uploaded media.
- Tamper-evident audit trails: Uses SHA-256 hashes and hash-chained records to support evidence integrity.
- Evidence review: Helps analysts inspect findings before finalising reports.
- Human-in-the-loop verification: Keeps final report approval with an authorised reviewer.
- Agent integrations: Supports OAuth-scoped, read-only investigation tools.
My core design principle is simple: AI should help people examine evidence, not replace their judgment.
An AI assessment is not proof of wrongdoing, and the platform is designed to support investigation rather than issue verdicts about people.
Demo
🔗 Live Project: https://suryakavacham.chitrakalpan.in/
The website includes a read-only demonstration corpus and a walkthrough of the investigation workflow, from media ingestion and analysis to human review and report sealing.
Live case data requires an analyst account.
Partner Technologies
Suryakavacham is built around a modular forensic workflow, secure evidence handling, and AI-assisted analysis.
Its architecture includes a metered AI gateway, validated inputs, independent analysis modules, and scoped agent integrations.
I will document the specific MLH partner technologies used in the implementation once they are verified against the active project code. I do not want to claim a partner integration that I cannot demonstrate.
Hackathon Experience
Building Suryakavacham has been an opportunity to explore a challenging question at the intersection of AI, digital media, cybersecurity, and trust.
The project pushed me to think beyond model predictions and consider evidence provenance, uncertainty, privacy, auditability, and the role of human reviewers.
I wanted to build something that treats responsible AI as an engineering requirement, not just a feature.
My biggest takeaway: In forensic AI, explaining what the system knows, what it does not know, and why a human must review the result is just as important as the analysis itself.
Event attended: [Add the official MLH hackathon name and your experience participating in it.]
Top comments (0)