Model exploration
Separate notebooks support experimentation with four transformer architectures for summarization.
Automatic Meeting Summarizer
From a long conversation to a concise summary.
Explore the repository ↗Why this project?
Meeting transcripts contain useful information, but their length can make the key points difficult to revisit. This project explores generating a shorter summary from a supplied text transcript.
From input to output.
The repository includes fine-tuning notebooks for T5, LED, BART, and Pegasus. Its Flask application uses a locally loaded fine-tuned Pegasus model to generate summaries through a web interface.
01Text transcript
02Whitespace cleanup
03Pegasus generation
04Summary
A simplified overview of the repository’s application flow.
Web interface
A Flask endpoint receives transcript text and returns the generated summary as JSON.
Inference pipeline
The application uses PyTorch, selects CUDA when available, and applies generation settings to control length and repetition.