Memoria — Overview
The Spark
(summary converted to Markdown paragraph)
The Science
We model the user's Cognitive Engagement Score (E_c) as:
$$ E_c = \frac{\sum_{i=1}^{n} \bigl(w_r \cdot R_i + w_e \cdot \varepsilon_i\bigr)}{T_{\text{response}}} $$
Where:
- (R_i) — accuracy of a recall event
- (\varepsilon_i) — emotional sentiment score (audio)
- (T_{\text{response}}) — response latency normalized to baseline
How We Built It
- Voice: ElevenLabs (streaming, low-latency)
- Brain: Google Vertex AI (Gemini Pro, large context)
- Infrastructure: real-time transcription → cognitive marker analysis → model-driven voice responses
Challenges
- Latency vs. empathy (optimized via low-latency WebSocket)
- Emotional nuance (prompt tuning to validate, not correct)
Built With
- ai
- api
- elevenlabs
- elevenlabs-react-sdk-(@elevenlabs/react)-architecture:-websocket-streaming-for-real-time
- firestore
- functions
- gemini
- gemini-pro
- google-cloud-functions
- google-firestore-apis:-elevenlabs-conversational-agents-api
- javascript
- node.js
- node.js-frameworks:-react-(vite)
- react
- tailwindcss
- tailwindcss-cloud-services:-google-cloud-vertex-ai-(gemini-pro)
- typescript
- vertex
- vite
Log in or sign up for Devpost to join the conversation.