MoodMetrics

Tracks audience emotion and attention during presentations using webcam data. After a session, aggregates everyone logs and shows a dashboard with engagement metrics and Claude-powered coaching tips.

Built at a hackathon, then refactored. See POSTMORTEM.md for what broke the first time and why.

---

How it works

Each presentation gets a session code you put on your title slide (e.g. CS4681-A). Audience members enter it in the GUI when they start recording. After you finish presenting, you pull their data and view it in the dashboard by session.

Pipeline: audience webcam -> emotion_logger_fixed.py -> sessions/{code}/emotion_{name}.jsonl -> pushed to GitHub -> server.js pulls and aggregates on demand -> React dashboard

---

Requirements

Presenter machine

• Node.js 18+
• npm install
• ANTHROPIC_API_KEY in .env
• No Python pip packages needed, aggregate_fixed.py is pure stdlib

Audience machines

• Python 3.10+ from python.org, check Add to PATH during install
• Git from git-scm.com, default options are fine
• Everything else installed automatically by audience_setup.bat

---

Setup

cp .env.example .env
# add your ANTHROPIC_API_KEY
npm install

For audience distribution: open audience_setup.bat and fill in your GITHUB_TOKEN before sending it out.

---

Running a presentation

Put your session code on your title slide. Keep it short like CS4681-A or honeywell-demo.

Audience steps during your intro:

1. Double-click audience_setup.bat
2. Enter their name and the session code
3. Click Start Recording
4. Click Stop and Submit when you finish

After presenting, open two terminals:

# terminal 1
node server.js

# terminal 2
npm run dev
# open http://localhost:3000

Pick your session from the dashboard, hit Pull Latest to grab the data, then Get AI Coaching Tips.

---

Testing locally without a webcam

Save this as seed_test.py and run it once, then start the servers normally.

import random, time, json, os
session = "test-session"
os.makedirs(f"sessions/{session}", exist_ok=True)
emotions = ["happy", "neutral", "sad", "angry", "fear", "surprised"]
start_ms = int(time.time() * 1000)
with open(f"sessions/{session}/emotion_aditya.jsonl", "w") as f:
    for i in range(100):
        top = random.choice(emotions)
        f.write(json.dumps({
            "timestamp": start_ms + i * 2000,
            "session": session, "user": "aditya",
            "distracted": random.random() < 0.15,
            "pitch": round(random.uniform(-25, 10), 2),
            "emotion": top,
            "top_emotions": [{"emotion": top, "score": round(random.uniform(0.4, 0.8), 4)}],
        }) + "

")

node server.js    # terminal 1
npm run dev       # terminal 2

---

Project structure

MoodMetrics/
- src/
  - App.tsx
  - main.tsx
  - index.css
  - components/
    - Dashboard.tsx
- sessions/                      created at runtime, gitignored
  - {session_code}/
    - emotion_{name}.jsonl       pushed by each audience member
    - summary.json               generated by server on demand
- audience_setup.bat             distribute this to audience
- audience_gui.py
- emotion_logger_fixed.py
- aggregate_fixed.py             pure stdlib, no pip install needed
- requirements_client.txt        audience machines only
- server.js
- vite.config.ts
- index.html
- package.json
- .env.example
- POSTMORTEM.md
- README.md

---

What the dashboard shows

• Session picker, one button per presentation
• Engagement score, distraction rate, top emotion, flagged moment count
• Emotion timeline chart in 30-second buckets
• Flagged moments where confusion or distraction spiked
• Audience members tracked per session
• AI coaching tips from Claude