SentinelAI

Sports IP Piracy Stops Here.

Every second, stolen sports footage is re-uploaded, re-encoded, and monetized without rights holders ever knowing. SentinelAI generates cryptographic DNA for every asset and runs a continuous AI cascade to detect every mutation — exact reposts, crops, overlays, screen captures — before they go viral.

Protect Your Content →Live Demo

Piracy Moves at Machine Speed

Sports broadcasters lose an estimated $28 billion annually to digital piracy. A final highlight lands online and within minutes it exists on a dozen platforms in mutated forms — cropped, re-encoded, screen-captured with watermarks burned off. Traditional takedown workflows are manual, slow, and always one step behind.

Rights holders need detection that matches the speed of infringement. SentinelAI matches it — and exceeds it.

exact repost1:1 re-upload, identical bytes
cropped or reencodedResolution, bitrate, aspect ratio changes
overlay or meme editCaptions, stickers, watermarks burned in
screen recorded recaptureScreen capture with glare artifacts
audio or semantic reuseSame commentary, different visual encoding

Digital DNA in Four Stages

Upload a video. SentinelAI builds a multi-layer fingerprint before the clip ever goes public — so any future copy can be matched against an immutable record.

01 // SHA-256

Source Hash

Cryptographic fingerprint of raw bytes — the immutable ground-truth anchor for every scan.

02 // SynthID

AI Watermark Token

Google SynthID-style token bound to asset identity — survives compression and re-encoding.

03 // dHash + Gemini

Keyframes + Scene Context

Keyframes extracted at scene changes, perceptually hashed via dHash, then Gemini annotates each shot — entities, actions, visual context — for rich frame-level intelligence.

04 // Gemini + Audio

Semantic Passport

Gemini generates the full content passport: scene-by-scene transcript, audio analysis, and semantic tags — all embedded into Pinecone for cross-mutation search.

Three-Stage Cascade — Fast to Deep

Ordered pipeline — lightweight checks run first, Gemini is called only when needed. Every decision is logged with confidence scores and a full audit trail.

// STAGE A — INSTANT

dHash + SynthID

Hamming distance under 10 bits → instant confirmed match. SynthID token lookup catches exact reposts in milliseconds.

// STAGE B — SEMANTIC

Semantic Embeddings

Gemini Embedding 2 (3072-dim) + Pinecone cosine similarity. Catches cropped, re-encoded, and overlay mutations that fool perceptual hashes.

// STAGE C — AI REASONING

Gemini Reasoning

Gemini Flash describes the suspect clip and compares it to the content passport. Handles screen captures, audio-only reuse, and adversarial edits.

Provenance Graph

Every confirmed match writes to a Neo4j Aura cloud graph. Nodes represent original assets, pirated clips, and platforms. Edges carry confidence scores and mutation labels — exposing the full distribution tree of illegal republishing.

Gemini analyzes each graph cluster for risk level and recommends automated action: DMCA notice, escalation, or monitoring. One click generates a pre-filled takedown email.

Asset NodePirated Clip NodePlatform Domain NodeConfidence Edge
Gemini Flash

Content Passport generation, Stage C reasoning, graph risk analysis

Gemini Embedding 2

3072-dimensional semantic vectors for cross-mutation similarity search

Google SynthID

AI watermarking for provenance chain — survives re-encoding

Pinecone on GCP

Vector index for sub-second semantic search at scale

Cloud Run + GCS

Serverless deployment; all video assets stored in Firebase Storage

Neo4j Aura

Cloud graph DB for violation relationship mapping