Breefli

Problem

In today's fast-paced environments, keeping track of important details from meetings, classes, and doctor visits is challenging. Manual note-taking often leads to missing critical points.

Solution: Breefly

Breefly is designed to summarize meetings, appointments, and other important events, helping users stay focused without multitasking.

Key Features:

Audio recording and transcription
Summarization of content
User-friendly features:
- Tagging and filtering summaries
- Sharing with friends
- Fast API integration to Whisper and GPT for summarization

Proof of Concept

Breefly hosts its own large language model (LLM) research project using the LLama 3-1_8b-instruct model. The team optimized for efficiency using NVIDIA A100 (80GiB of parameters).

Fine-tuning Process:

Dataset split into training, validation, and testing sets.
Fine-tuning done by multiplying matrix weights of the LLM with the dataset, resulting in a "fine-tuned" model.