AAC Prompt Classifier

A Streamlit web app that analyzes Augmentative and Alternative Communication (AAC) letterboarding therapy session transcripts. Upload an SRT file and the app classifies every facilitator prompt into one of five evidence-based categories using GPT-4.

Prompt categories

Category	Description
Choice	Offers the student discrete, explicit options
Clarification	Asks the student to confirm or expand on something they already communicated
Guided	Scaffolds toward a specific answer via letter cues or directional hints
Open-ended	Invites a free response with no constraints or implied target
Reinforcement	Encourages effort or regulates emotion without eliciting new information

Project structure

.
├── app.py                        # Streamlit entry point
├── requirements.txt
├── .env.example                  # Copy to .env and fill in your keys
├── Makefile                      # Common dev tasks
├── pipelines/
│   └── steps/
│       ├── classifier.py         # OpenAI GPT-4 classification
│       ├── database.py           # Supabase persistence
│       ├── ingest_transcript.py  # SRT parsing
│       └── metrics.py            # Aggregation helpers
├── templates/
│   └── report_template.html      # Embedded HTML report
├── tests/
│   └── test_pipeline.py
└── data/                         # Sample / training data

Setup

1. Prerequisites

• Python 3.11+
• An OpenAI API key
• A Supabase project with a sessions table

2. Install dependencies

python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

3. Configure environment variables

cp .env.example .env
# Edit .env and fill in OPENAI_API_KEY, SUPABASE_URL, SUPABASE_KEY

4. Run the app

streamlit run app.py
# or: make run

Usage

1. Open the app in your browser (default: http://localhost:8501).
2. Upload an .srt transcript file.
3. Click Analyze.
4. Review the prompt distribution chart and the classified prompt table.

Development

make test    # Run unit tests
make lint    # Lint with flake8

Database schema

The sessions table in Supabase should have at minimum:

create table sessions (
  id          uuid primary key default gen_random_uuid(),
  file_name   text,
  results     jsonb,
  created_at  timestamptz default now()
);