Observe
Capture multi-turn intent, affect, safety context, and expertise signals.
3
user modes
3
memory layers
TXT/PDF/DOCX
personal context uploads
0
real payments in Figure 1
Demo thesis
A twin should be trained by interaction.
AI YOU treats the user as an evolving source of evidence. Each conversation turn, correction, upload, and feedback event can update the twin's memory and reward traces while preserving a small, auditable record for later learning.
Personalize
Use mode-aware memory and uploaded knowledge to shape the next response.
Reward
Convert ratings, corrections, freeze marks, and redactions into feedback data.
Replay
Keep privacy-aware trajectory summaries available for offline improvement.
Figure 1 marketplace / Lily money market
A deterministic trade chain for recording the core demo.
The marketplace route starts with Riverside High hiring Lily for youth wellbeing and puberty guidance. The screen spends simulated credits, produces a guidance artifact, unlocks recommended next trades, and records an activity feed.
AI YOU Town Marketplace
simulated settlement only
buyer
Riverside High
limited funding scenario
twin
Lily
puberty guidance session
unlock
Parent Q&A pack
recommended next trade
Artifact
Student guidance pack
Notes for body changes, private questions, consent, and escalation.
Activity feed
Order -> artifact -> recommendation
Every state change is visible for a clean paper-video capture.
Safety boundary
No real money
Demo credits model marketplace behavior without a payment provider.
System architecture
Static site in front, live system on the Docker Space.
GitHub Pages presents the project narrative. The working demo stays on Hugging Face because AI YOU needs FastAPI, WebSockets, uploads, memory state, and feedback endpoints.
-
01
React playground
Persona lobby, mode-gated chat, twin workspace, upload controls, and Figure 1 route.
-
02
FastAPI + WebSocket API
Session start, chat stream, uploads, feedback, trajectory, and marketplace endpoints.
-
03
Mode-aware orchestrator
Prediction, emotion, safety, memory, relationship, and twin modules run by mode contract.
-
04
Learning layer
Privacy-aware trajectory storage, reward decomposition, and offline improvement hooks.
Demo signals / metrics
Measure what the audience can actually inspect.
Signal
Where it appears
Why it matters
Mode contract
Consultation, companion, and normal modes
Shows that hidden modules do not keep running outside the selected mode.
Memory depth
Working, episodic, and semantic summaries
Makes the personal twin more than a one-shot prompt.
Marketplace state
Artifacts, recommendations, and activity feed
Gives Figure 1 a repeatable visual path from order to next trade.
Feedback data
Ratings, corrections, redactions, and reward summaries
Connects human judgment to future training artifacts.
Recording flow
The shortest clean path for the paper demo video.
Start from the live marketplace route, reset the state, and record the visible trade chain without relying on hidden setup.
Open the marketplace route- Open /marketplace on the Hugging Face Space.
- Click Reset demo for a fresh Riverside High scenario.
- Hire Lily with simulated credits and wait for the guidance artifact.
- Show the recommended next trades that unlock from the artifact.
- Close on the activity feed so the order, artifact, and recommendation are visible.
Source links