CogPace

MFA attention fields power adaptive tutoring — on-device Arm64 inference, Python Streamlit core, GitLab Duo-ready agent telemetry.

Comment

Inspiration

STEM and dev tools measure outcomes, not the attention collapse before failure. CogPace models tutoring with the Magnetic Field Model of Attention (MFA): field strength S, load E_total, and peripheral capture (Lavie, 1995; Posner, 1980). States export as JSON telemetry for humans and agents.

What it does

  • Streamlit tutor with live MFA states (OPTIMAL → OVERLOADED)
  • Zero-preset adaptation — difficulty follows attention equations, not Python char→answer tables
  • cora.zone demo + GitHub open source

How we built it

Layer Stack
Core Python · Streamlit · mfa_attention.py
Links https://github.com/xqscora/cogpace · https://cora.zone
Video YouTube demo on project page

Challenges

Making attention a first-class signal (not UX polish) while keeping adaptation emergent from MFA parameters (sigma, F_thresh).

What's next

On-device inference (Arm), GitLab Duo agent workflows, Qwen Cloud memory agents for cross-session tutoring.

Built With

  • arm64
  • edge-inference
  • mfa-attention-model
  • on-device-ai
  • python
  • streamlit
Share this project:

Updates