Back to Projects
JARVIS: Local AI Personality Clone
Engineered a privacy-first AI assistant running 100% locally with no cloud APIs, using Ollama (Llama 3), ChromaDB, and custom RAG pipelines to mirror personal writing styles.
Talk about a similar project
About
Built a fully on-device, privacy-centric AI assistant that acts as a personal clone with no cloud APIs. The system is trained on writing samples to learn tone, vocabulary, and style, using Ollama (Llama 3) for local inference and ChromaDB for semantic vector memory. It features a custom NLP analysis engine, RAG-based memory, self-scoring response quality, and voice input/output through a Gradio UI with OpenAI Whisper and Edge TTS. Released as open-source software.
Key Features
- Privacy-first, 100% local inference with no cloud APIs using Ollama and Llama 3
- Semantic RAG memory system powered by ChromaDB
- Dynamic personality cloning pipeline trained on writing samples to learn tone, vocabulary, and style
- Self-scoring response quality with voice input/output using Whisper STT and Edge TTS
Info
Tech
PythonOllama (Llama 3)LangChainChromaDBRAGsentence-transformersWhisperEdge TTSGradio
