mohamedabbouda/confidence-qa-demo

📝 Confidence-Based Document QA Extraction with Feedback Logging and Web Demo

This project is a prototype question answering (QA) system built as part of an internship technical task. It uses a pre-trained Hugging Face extractive QA model to answer questions from document-like text. The system includes confidence-based filtering, feedback logging, and an interactive Gradio web app.

🚀 Features

• —Extracts answers from text using a pre-trained BERT QA model
• —Returns answer + confidence score
• —Flags low-confidence predictions (default threshold = 0.75)
• —Saves flagged outputs to feedback_log.jsonl
• —Allows manual corrections → stored in corrected_data.jsonl
• —Web interface (Gradio) for interactive testing and corrections

📂 Project Structure

confidence_qa/
├── app.py # Web app (Gradio)
├── qa_pipeline.py # Backend QA pipeline with threshold + logging
├── test_model.py # Quick test script
├── requirements.txt # Dependencies
├── feedback_log.jsonl # Auto-generated flagged outputs
├── corrected_data.jsonl # Manually corrected answers
└── README.md # Project documentation

⚙️ Installation & Setup

1. 1.Clone the repo:

   git clone https://github.com/alex-dev/confidence_qa.git
   cd confidence_qa

1. 1.Create and activate virtual environment:

python -m venv .venv
source .venv/bin/activate   # macOS/Linux 

1. 1.Install dependencies:

 pip install -r requirements.txt

⚙️ Run pipeline (generate feedback logs)

 python qa_pipeline.py

⚙️ Run web app

python app.py

⚙️ 📝 Example

Context:

The Eiffel Tower is in Paris. It was completed in 1889.

Question:

When was the Eiffel Tower completed?

Output:

Answer: 1889
Confidence: 0.92
Flagged: False