VeriMatChain AI
Multimodal verification platform powered by artificial intelligence designed to combat misinformation and synthetic content through semantic analysis, RAG fact-checking, decentralized storage on IPFS, and immutable attestations on Solana. The system analyzes text, URLs, and audio through a deterministic pipeline with SHA-256 hashing, contextual reasoning with Claude AI, and cryptographically verifiable evidence.
The modern digital ecosystem faces an unprecedented informational integrity crisis: generative AI enables misinformation at industrial scale, synthetic audio and voice cloning are indistinguishable from the original, and traditional verification systems lack transparency, immutability, and explainability. No open infrastructure exists that combines semantic reasoning with cryptographically verifiable and permanent evidence.
VeriMatChain AI implements a six-stage deterministic verification pipeline on LangGraph: content normalization and SHA-256 hashing, contextual evidence retrieval via RAG with Google Fact Check Tools API, structured semantic analysis with Claude AI evaluating bias, manipulative rhetoric, and unverifiable claims, decentralized preservation of complete evidence on IPFS via Pinata, and immutable attestation on Solana Devnet using Program Derived Addresses. Each verification produces a cryptographically auditable chain of custody.
- 01
Designing a LangGraph pipeline where each node operates as an independent unit with single responsibility, enabling complete observability and extensibility without breaking the deterministic flow
- 02
Implementing deterministic SHA-256 hashing over URL-extracted content with BeautifulSoup, removing scripts, ads, and irrelevant markup to preserve only semantically relevant content before fingerprinting
- 03
Building a Solana attestation system using Program Derived Addresses that minimizes on-chain storage by storing only irreducible proof artifacts while maintaining complete evidence off-chain on IPFS
- 04
Integrating ElevenLabs AI Speech Classifier for synthetic audio detection as an additional pipeline layer without introducing significant latency into the main text and URL verification flow
- 05
Calibrating the integrity scoring model from 0.0 to 1.0 to function as an assistive signal rather than an absolute truth engine, explicitly communicating system limitations in the interface
- 01
Hybrid on-chain/off-chain architecture is not a compromise but a deliberate design decision: heavy AI operations remain off-chain to reduce costs and latency, while cryptographic proof artifacts go on-chain to guarantee long-term auditability
- 02
LangGraph as an AI pipeline orchestrator is superior to manual sequential chaining because each node can be observed, restarted, and replaced independently without affecting the global pipeline state
- 03
Claude produces more useful semantic analysis when the prompt explicitly specifies the dimensions to evaluate: framing bias, manipulative rhetoric, unattributed claims, and contextual consistency, rather than asking for a generic truthfulness classification
- 04
Solana Devnet with PDAs is a production-ready attestation architecture: costs are extremely low, throughput is sufficient for real-time verification, and deterministic address derivation eliminates the need for centralized indexes