Skip to content

About

Realtime LM Dashboard for Intervention Studies

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Learner Modeling Dashboard

The Learner Modeling Dashboard is a live "who needs help" board for a room of students who code in VEX. The dashboard does these tasks:

  • It copies the activity of the students from the Reflecks production backend to your own machine.
  • It measures how much the code of each student changes between runs. The measure is a plain edit distance, with no black-box model.
  • It divides the session into episodes.
  • It shows on one screen who needs attention now (wheel-spinning, resilience, inactive, explorer, step-by-step).

The dashboard only reads from production.

flowchart LR
    students["Students coding<br/>in VEX"] --> prod[("Reflecks<br/>production server")]
    prod -. "polls, read-only" .-> daemon["Local daemon<br/>mirror and analyze"]
    daemon --> pg[("Postgres<br/>mirror")]
    pg --> api["Read API"]
    api --> dash["Researcher<br/>dashboard"]
Loading

The full documentation is at https://inviteinstitute.github.io/lm-dashboard/. To read it locally on port 4000, run mkdocs serve.

Quick Start

You need only Docker: Docker Desktop on Mac or Windows, or Docker and the compose plugin on Linux. One compose file runs the whole stack, the same on your laptop and in prod. The stack has these parts:

  • Postgres
  • The API, which serves the built dashboard
  • The ingestion daemon
cp .env.example .env.mirror   # fill in PROD_USERNAME / PROD_PASSWORD
echo "POSTGRES_PASSWORD=$(python3 -c 'import secrets;print(secrets.token_hex(16))')" > .env
make dev                      # or: docker compose up -d

make (or make help) lists the shared commands: dev, test, lint, format, build and deploy. See the Development docs.

  1. Open the dashboard.
  2. Sign in with PROD_USERNAME and PROD_PASSWORD.
  3. Add a student ID.

The data for that student starts to arrive while the board is open.

The Quickstart docs have the full setup: dev and prod, configuration, and accounts.

What You Get

  • A card for each student. The card has a run track, an episode sparkline, a status badge from the active triggers of the student, and Present and Picked toggles. The run track has one tile for each run. The color of a tile shows how many blocks changed since the last run.
  • A live "who needs help" column. Five edit-distance triggers (wheel-spinning, resilience, inactive, explorer, step-by-step) drive this column. You can write notes on each alert. Click a learner to see the full detail and notes log.
  • A top bar. Use it to pause and resume polling, to export all data as a zip of CSV snapshots, and to reset the board.

Serving It Remotely

Run the prod stack (docker compose -f compose.yml up -d). The API binds to 127.0.0.1:8000, so nginx or another reverse proxy sits in front of it. HTTP Basic (the native login dialog of your browser) gates the whole origin and checks against the researcher table. Thus nothing is ever exposed without the gate. While the app is served, a dead-man's switch pauses the production polling of a board when its dashboard is not open.

Under the Hood

The daemon is the only process that writes. The API and dashboard only read a Postgres store. You can rebuild this store from the raw event log at any time. The full documentation covers configuration, the API and the architecture. It is at https://inviteinstitute.github.io/lm-dashboard/.

About

Realtime LM Dashboard for Intervention Studies

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Packages

Contributors

Languages