-
-
The AI Autopsy: Pinpointing logic drift and ethical runoff in real-time.
Inspiration
As AI agents evolve from simple chatbots into autonomous decision-makers, a terrifying reality is emerging: the "Black Box" problem. If an AI suggests tax evasion or ignores critical stroke symptoms, it's not just a software bug—it's an ethical crisis. We realized developers lack the tools to investigate why a model failed. We built Synapse Sentinel because the AI revolution needs a forensic audit trail.
What it does
Synapse Sentinel is a real-time investigator's dashboard. It intercepts high-risk AI interactions and performs an instant "AI Autopsy."
- Real-Time Risk Scoring: It assigns a critical severity score based on Logic Drift and Ethical Runoff.
- Fracture Point Detection: It pinpoints the exact tokens in the LLM's reasoning where safety heuristics were bypassed.
- Ethical Patching: It instantly generates an optimized system directive to repair the model's behavior and files a compliance report.
How we built it
We leveraged Google Gemini 1.5 Pro as our core intelligence engine to power the root-cause analysis and generate the ethical patches. For the frontend, we built a highly immersive "Cyber-Diagnostic" HUD using Next.js and Tailwind CSS. We utilized Framer Motion to create dynamic glitch effects and fluid state transitions, giving the dashboard a high-stakes, real-time tactical feel.
Challenges we ran into
Quantifying abstract concepts like "Ethical Runoff" into a hard 0-100 Risk Score was difficult. We had to carefully engineer our prompts to ensure the diagnostic engine could consistently evaluate fintech, medical, and commerce cases without bias. Additionally, managing the complex React states to ensure the UI transitions smoothly during the "Scanning" phases required precision engineering.
Accomplishments that we're proud of
We successfully built a tool that doesn't just say "The AI is wrong." It provides actionable, forensic evidence of where it went wrong and how to fix it. The seamless integration of our "Project Brief" modal and the dynamic risk gauge makes this prototype feel like a fully-fledged enterprise product.
What's next for Synapse Sentinel
We plan to integrate live API hooks for enterprise LLMs (like OpenAI and Anthropic) so that Sentinel can act as an active, in-line firewall—blocking toxic or dangerous outputs before they ever reach the end user's screen.
Built With
- framer-motion
- google-gemini
- html
- next.js
- react
- tailwind-css
- typescript
