Sententia
2025archivedAI-assisted learning prototype connecting Raspberry Pi hardware to a Gemini-powered reasoning loop — active, feedback-driven learning instead of passive instruction.
Story
Most "AI security camera" demos pretend the camera does the thinking. In reality, the AI inference is heavy and the camera is a Raspberry Pi. Sententia separates the roles honestly: edge devices stream frames, a central brain on a laptop does the reasoning, and the only thing on the wire is structured JSON.
The hard part was deciding which frames to sample. Run inference on every frame and you blow the API budget in minutes. Sample naïvely and you miss the moments that matter. Sententia uses intelligent frame selection driven by simple motion deltas before paying for an AI round trip.
What I built
- 1Built the brain-edge architecture separating AI inference (laptop) from camera streaming (Raspberry Pi)
- 2Implemented real-time intent classification with intelligent frame sampling for API cost management
- 3Designed a Streamlit monitoring dashboard with live threat visualization and history