How zackriya-solutions/meeting-minutes works
Project Type & Purpose
• Recognize that this is a full-stack application designed for capturing live meeting audio, transcribing it in real-time, and generating summaries via AI.
• The solution emphasizes local processing (privacy-first), cost effectiveness, and support for multiple LLM providers as specified in the README.
• That the project involves components built with multiple languages and frameworks (Python, Rust, Next.js/Tauri, and C++).
Analyzing the File Structure
• Top-Level Directories:
– backend: Contains Python scripts (e.g., main.py, Process_transcrip.py, db.py) to run FastAPI servers, API documentation, shell scripts to manage processes, and a “whisper-custom” folder with C++ code (server.cpp) and supporting files indicating a custom transcription engine.
– frontend: Includes Next.js configuration, Tauri-specific sources (Rust code under src-tauri), and supporting scripts, indicating the desktop application and web UI aspects.
• Configuration and build files (e.g., package.json, requirements.txt, CMakeLists.txt) show the dependency management and build orchestration.
Extracting Additional Insights from the README
• The README explicitly outlines the high-level architectural components:
– Audio Capture Service: Built partly in Rust (experimental) and Python for capturing real-time audio.
– Transcription Engine: Uses Whisper.cpp for local transcription.
– LLM Orchestrator: Manages integration with external LLM providers (Anthropic, Groq, Ollama) with fallback mechanisms.
– Data Services: Includes ChromaDB for vector embeddings and SQLite for metadata storage.
– API Layer: Built with FastAPI to provide endpoints (upload, process, summary retrieval, deletion).
– Deployment Architecture: Frontend built as a Tauri app coupled with Next.js and backend with Python/FastAPI transcript workers.
• This detailed description gives you the logical components and their interactions.
a. Core Components:
• Frontend Component:
• Next.js/Tauri (Desktop and UI layer)
• Client-side components (meeting notes, audio controls, etc. from src/app and src/components)
• Backend Component:
• API Layer (FastAPI endpoints, process orchestration)
• Audio Capture Service (scripts, audio processing modules in Python/Rust)
• Transcription Engine (Whisper.cpp integration; C++ and Python scripts under whisper-custom and backend/app)
• LLM Orchestrator (manages interactions with Anthropic, Groq, and Ollama APIs)
• Data Services:
• Database(s): Likely SQLite for process tracking and possibly ChromaDB for transcript embeddings.
• External Dependencies & Configurations:
• External libraries like FFmpeg, Node.js, Python, Rust, and C++ compilers.
• Environment variables and API keys as indicated in the README (.env file requirements).
b. Relationships and Data Flow:
• Audio flow from the Audio Capture Service to the Transcription Engine.
• Transcription output flowing to the LLM Orchestrator for summarization.
• Client requests from the Frontend interacting with the Backend API Layer.
• Data movement between the Transcription Engine/LLM processes and Data Services for storage/retrieval.
• How the API layer orchestrates these services through endpoints (upload, process, summary).
c. Architectural Patterns and Design Principles:
• Separation of Concerns: Clearly separate presentation (frontend) from business logic (backend) and data storage (SQLite/ChromaDB).
• Modular Design: Independent components (audio capture, transcription, LLM processing) communicate via clearly defined APIs.
• Local Processing Emphasis: Privacy-first design where core AI processing happens locally.
d. Technologies/Frameworks:
• Frontend: Tauri (Rust), Next.js (React/TypeScript), Tailwind CSS.
• Backend: Python (FastAPI), Whisper.cpp, custom C++ components, shell scripts.
• Others: Rust for experimental audio capture; ChromaDB and SQLite for data management.
Final Recommendations
• Be as detailed as possible – don’t shy away from splitting the backend into multiple components (e.g., separate audio capture, transcription, and LLM orchestrator blocks).