How aitrekker/jarvis works
Project Type & Purpose
• Jarvis is a full-stack, modular AI voice-intelligence platform that captures live audio, transcribes it, summarizes content, indexes it semantically, and exposes both CLI and web UIs for retrieval and exploration.
Main Architectural Components
a. Audio Capture Module
• utils/recorder.py + debug/ui.py + platform FFmpeg binary
• Streams microphone input, preprocesses audio
b. Transcription Engine
• OpenAI Whisper integration (search/ollama_helper.py)
• search/search.py, search_engine.py for managing transcription jobs
c. Summarization Pipeline
• utils/summarize.py + Ollama phi4 model for local LLM summarization
• utils/periodic_tasks.py for scheduled summary jobs
d. Persistence Layer
• FileStore (storage/file_store.py) for raw transcripts
• ChromaDB vector store (storage/chroma_store.py, storage/chroma/) for embeddings & summaries
• storage/models.py defines metadata schemas
e. Semantic Search Engine
• search/search_engine.py + search_cli.py
• Embedding generation via nomic-embed-text
f. User Interfaces
• CLI: start_Jarvis.py, search/search_cli.py
• Web UI: Streamlit frontend (web/Jarvis_UI.py + web/components + web/web_utils)
g. Configuration & Logging
• config.py for global settings
• setup/logger.py + debug/logging.py for multi-level logs & diagnostics
h. Orchestration & Setup
• setup_Jarvis.py / setup/setup.py for bootstrap & env setup
• start_Jarvis.sh wrapper
Component Interactions & Data Flow
1. Microphone → Audio Capture Module → (FFmpeg) → Raw Audio Buffer
2. Buffer → Transcription Engine → Transcript Text
3. Transcript → FileStore → Raw transcript files
4. Batch transcripts → Summarization Pipeline → Summary Text
5. Summary + Transcript → Embedding Generator → Vector embeddings
6. Embeddings → ChromaDB → Indexed for similarity search
7. CLI / Web UI → Semantic Search Engine → Query ChromaDB → Results
8. Periodic Tasks Scheduler → triggers summarization & cleanup jobs
Architectural Patterns & Principles
• Clear separation of concerns: capture, processing, storage, UI
• Pipeline pattern: sequential processing stages (transcribe → summarize → index)
• Modular services: loosely coupled Python modules for easy extension
• Layered architecture: UI layer, service layer (transcribe, summarize, search), data layer
Technologies & Frameworks
• Python 3.11, FFmpeg (binaries/)
• OpenAI Whisper (speech-to-text)
• Ollama LLM (phi4) for summarization
• nomic-embed-text for embeddings
• ChromaDB for vector storage
• Streamlit for web UI
• pytest for automated testing