A browser extension that gives you a sixth sense for trust.
Sentinel is a Chrome extension that helps users instantly understand the trustworthiness of anything they see online — posts, profiles, images, comments, and DMs.
Instead of simply detecting “AI-generated content,” Sentinel interprets patterns across the web using:
- Actian Vector Database → memory, similarity search, pattern clustering
- Sphinx Reasoning Engine → human-readable explanations, intent detection, risk interpretation
- Sentinel doesn’t tell you what’s true.It tells you what’s trustworthy — and why.
The modern internet is full of:
- Fake accounts
- Recycled misinformation
- Emotionally manipulative posts
- AI-generated images
- Scam DMs
- Coordinated bot networks
- Most people can’t tell what’s real anymore.
- Sentinel solves the real problem:
- “I don’t know what I can trust.”
- It gives users a new sense — a trust intuition powered by AI.
🔹 1. Passive Trust Signals (Automatic)
As you scroll, Sentinel automatically analyzes each post and injects a small Trust Signal beneath it.
Each Trust Signal includes:
- Trust Score
- Pattern Matches (via Actian)
- Risk Indicators (manipulation, impersonation, bot-like behavior)
- Sphinx Reasoning Summary (“why this matters”)
This happens in real time with zero user effort.
🔹 2. Deep Check (Manual X‑Ray Mode)
Right‑click any image, profile, or text → Deep Check with Sentinel
Deep Check reveals:
- Similarity to known patterns
- Cluster membership (bot networks, repeated images)
- Emotional manipulation cues
- Contradiction detection
- Intent analysis
- Safety risks
This is your X‑ray vision for the internet.
🔹 3. Pattern Memory (Actian Vector DB)
Sentinel stores embeddings of:
- Suspicious images
- Bot-like profiles
- Repeated claims
- Scam patterns
- Manipulative language
This enables:
- “You’ve seen this pattern before” alerts
- Detection of recycled scams
- Cluster detection
- Contradiction detection
- Long-term trust intelligence
🔹 4. Sphinx Reasoning Engine
Sphinx turns raw signals into clear, human-readable explanations, such as:
- “This account resembles a bot cluster you encountered earlier.”
- “This image is 92% similar to a known AI-generated pattern.”
- “This DM contains coercive language patterns.”
- “This claim contradicts a post you saw yesterday.”
This is what makes Sentinel feel like a new sense, not a tool.
Frontend
- React + Vite
- Tailwind CSS
- Chrome Extension (Manifest V3)
- Content Scripts + DOM Injection
- MutationObserver for passive scanning
Backend
- FastAPI
- Actian VectorAI DB
- Sphinx Reasoning SDK
- CLIP Embeddings (HuggingFace)
This is the full end‑to‑end flow that powers Sentinel’s trust intelligence.
1. DOM Watcher (Frontend)
Sentinel monitors the page as you scroll.
Detect new posts/images/profiles
Extract:
- Image URLs
- Profile picture
- Username
- Bio
- Post text
- Send payload to backend
- Show “Analyzing…” skeleton loader
2. Embedding Engine (Backend)
FastAPI generates a CLIP embedding for the content.
3. Actian Vector Search (Backend)
Actian retrieves:
- Similar images
- Similar profiles
- Similar claims
- Cluster membership
- Past encounters
4. Sphinx Reasoning (Backend)
Sphinx interprets the signals and produces:
- Trust score
- Reasoning summary
- Risk indicators
- Intent analysis
- Contradiction detection
- Manipulation cues
5. Trust Signal Injection (Frontend)
Sentinel injects a small UI card under the content:
- Trust Score
- “Why this matters” explanation
- Similarity matches
- Risk indicators
- Deep Check button
6. Deep Check (Optional)
User triggers Deep Check via right‑click or button.
Backend performs:
- Deeper similarity search
- Cluster analysis
- Manipulation detection
- Contradiction detection
- Intent reasoning
- Frontend displays a detailed breakdown.
7. Pattern Memory Update (Backend)
- Actian stores new embeddings and updates clusters.
- Sentinel gets smarter with every page you browse.
- Scroll Twitter/Instagram → Trust Signals appear automatically
- Right‑click → Deep Check → X‑ray analysis
- “You’ve seen this pattern before” → Actian cluster reveal
- Toggle Trust Mode → page overlays highlight risks
(fill this in once packaged.)
Team Names