Automated Lab Book Generator V2

A Python application that records lab sessions, identifies speakers, and uses LLMs to generate structured lab books with multimodal support, lab cycle organization, and advanced analysis.

New Features in V2

• Lab Cycle Organization: Group related lab sessions into cycles (e.g., "Wave Experiments")
• RAG Knowledge Base: Each lab cycle builds a knowledge base for improved context in future sessions
• Enhanced Speaker Diarization: Better distinguish between the primary user and external speakers
• External Comments Section: Lab partner or demonstrator comments are properly categorized
• Post-Processing Analysis: Get feedback and improvement suggestions from advanced LLMs

Features

• Audio Recording: Captures lab sessions with automatic silence detection
• Speech Recognition: Transcribes audio using OpenAI's Whisper
• Speaker Diarization: Identifies different speakers in the recording
• Lab Book Generation: Creates structured lab books using a local LLM
• Multimodal Support: Analyzes graphs and images for inclusion in lab books
• Multiple Output Formats: Generates lab books in Markdown and/or DOCX format

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/labbook-generator.git
   cd labbook-generator

2. Run the setup script:

   chmod +x setup-v2.sh
   ./setup-v2.sh

3. (Optional) Set API keys for advanced post-processing:

   export OPENAI_API_KEY=your_openai_key_here
   export ANTHROPIC_API_KEY=your_anthropic_key_here

4. (Optional) Set up speaker diarization with HuggingFace:

   export HF_TOKEN=your_huggingface_token_here

Usage

Create a Lab Cycle

Organize your lab sessions into cycles:

python main_updated.py --create-cycle waves --cycle-title "Wave Experiments" --cycle-desc "Lab experiments focusing on wave properties and interference"

Record a Lab Session in a Cycle

python main_updated.py --record --cycle-id waves

This will start recording. Use the following commands during recording:

• p - Pause recording
• r - Resume recording
• s - Save current segment and start a new one
• l - Generate interim lab book
• q or Ctrl+C - End session and generate lab book

Process an Existing Audio File

python main_updated.py --process path/to/audio_file.wav --cycle-id waves

List Lab Cycles

python main_updated.py --cycles

List Sessions in a Cycle

python main_updated.py --sessions

Additional Options

• --output-format: Choose output format (markdown, docx, or both)
• --model: Specify a custom LLM model path
• --whisper-model: Select Whisper model size (tiny to large)
• --post-process: Enable analysis with advanced LLM APIs (default if API keys available)
• --no-post-process: Disable advanced LLM post-processing
• --api: Choose API provider for post-processing (openai or anthropic)

Example:

python main_updated.py --record --cycle-id waves --output-format markdown --whisper-model small --api anthropic

Lab Cycle RAG System

The RAG (Retrieval Augmented Generation) system automatically:

1. Stores all lab book content in a knowledge base for each lab cycle
2. Retrieves relevant context from previous lab sessions when generating new lab books
3. Provides continuity and knowledge transfer between related sessions

Enhanced Speaker Diarization

The system now:

1. Distinguishes between the primary user and external speakers
2. Creates a separate "External Comments" section in the lab book
3. Maintains profiles for known speakers for better identification

Post-Processing with Advanced LLMs

After generating a lab book, the system can:

1. Send the content to OpenAI or Anthropic APIs for advanced analysis
2. Receive feedback on strengths, weaknesses, and potential improvements
3. Get suggestions for follow-up experiments or additional analyses

Configuration

Edit config_updated.py to customize:

• Default lab cycle
• LLM parameters
• Speaker diarization settings
• RAG system configuration
• Post-processing options

Known Limitations

• Speaker diarization accuracy depends on audio quality and speaker separation
• RAG context retrieval requires sufficient previous lab content
• Advanced post-processing requires API keys and internet access

License

MIT License