Internship Project Proposal

Private AI

Self-Hostable Multilingual Audio Transcription and Summarization System for Enhanced Meeting Analysis.

Objective

The project aims to develop a self-hostable, private audio transcription system integrating state-of-the-art multilingual speech recognition models and large language models (LLMs). It will facilitate meeting analysis by enabling speaker identification, summarizing meeting transcripts, and contextualizing the summaries with meeting agendas.

Project Context and Justification

As companies engage in increasingly global operations, handling multilingual meetings has become crucial for effective collaboration. However, standard transcription services may not meet the specific needs of European companies focused on privacy, customization, and compliance with local regulations. Bitwise Lab, specializing in AI and data privacy, seeks to develop a robust solution that can be hosted privately, ensuring data confidentiality while leveraging advanced AI capabilities.

Project Goals

• Development of Multilingual Speech Recognition Models: Explore and evaluate various speech recognition models to transcribe conversations in different languages commonly used in EU workplaces.
• Speaker Identification and Separation: Implement features to identify and separate speakers for more accurate transcription of multi-participant conversations.
• LLM-based Summarization with Contextual Insights: Integrate LLMs to generate accurate meeting summaries, incorporating context from predefined agendas for tailored, action-oriented summaries.
• User Interface for Transcript Management: Design an intuitive, user-friendly interface for real-time transcript viewing, editing, and exporting.

Methodology

• Literature Review and Model Selection: Begin with a comprehensive review of current multilingual speech recognition models and LLMs suited for transcription and summarization.
• Development and Testing: Build a scalable, secure system using open-source and proprietary tools as needed, focusing on accuracy and multilingual capabilities.
• User Evaluation: Evaluate the system with end-users in simulated meeting scenarios to gather feedback for iterative improvement.

Expected Deliverables

• A functional prototype of the self-hostable transcription system.
• Documentation and final report detailing the technical methodology, challenges encountered, and future recommendations.
• Deployment guidance for private hosting.

Project Duration

5 months

Mentorship and Resources Provided by Bitwise Lab

The project will be supervised by experienced engineers at Bitwise Lab, providing technical support, computing resources, and guidance. This mentorship will ensure that the project meets both scientific and practical standards.

Funding Opportunities

EU students placed at Bitwise Lab may be eligible for the Erasmus+ Programme, supporting their stay throughout the project duration.

Conclusion

This project will allow students to gain hands-on experience in developing AI-powered applications within a privacy-centric context. It aligns with the strategic goals of Bitwise Lab in advancing AI and privacy, while addressing a growing need for secure, multilingual transcription solutions in the EU.