How afshazareen/clarity_project works
CLARITY analyzes political question-and-answer pairs, returning a clarity class and a finer-grained evasion technique. Users submit pairs through the browser UI to a Flask API; the API validates requests and delegates inference to a predictor that uses trained models when available and heuristic fallbacks otherwise. A separate training workflow loads QEvasion data and produces the two model artifacts. The README documents the UI, API endpoints, batch analysis, examples, and training workflow; the frontend source was not sampled, so its internal behavior is not detailed.
Main entry points: predict.py Predictor · train.py Training Pipeline · index.html Browser UI · app.py Flask API
Components 11
User Interaction
- Browser UI
frontend/index.html - Flask API
backend/app.py
Inference
- Predictor
backend/predict.py - Clarity Classification
backend/predict.py - Evasion Classification
backend/predict.py - Heuristic Fallbacks
backend/predict.py
Model Training
- Training Pipeline
backend/train.py - QEvasion Dataset
Model Artifacts
- Clarity Model
models/clarity_model/config.json - Evasion Model
models/evasion_model/config.json
Other components
- Analyst
Connections 11
- Analyst → Browser UI: submits pairs
README.md - Browser UI → Flask API: sends requests
README.md - Flask API → Predictor: requests predictions
backend/app.py - Predictor → Clarity Classification: classifies clarity
backend/predict.py - Predictor → Evasion Classification: classifies evasion
backend/predict.py - Predictor → Heuristic Fallbacks: uses when unavailable
backend/predict.py - Clarity Classification → Clarity Model: loads model
backend/predict.py - Evasion Classification → Evasion Model: loads model
backend/predict.py - Training Pipeline → QEvasion Dataset: loads data
backend/train.py - Training Pipeline → Clarity Model: trains and saves
backend/train.py - Training Pipeline → Evasion Model: trains and saves
backend/train.py