J.A.R.V.I.S. is a local-first personal AI assistant built in Python, designed to combine voice interaction, local LLM reasoning, persistent memory, RAG over study documents, and lightweight PC automation.
The project explores how a personal AI system can become context-aware without depending entirely on cloud infrastructure.
Most personal assistants are either simple chatbots or cloud-dependent automations.
J.A.R.V.I.S. takes a different approach:
- local reasoning with Ollama
- cloud fallback only when needed
- voice interaction through Alexa / speech interfaces
- persistent memory with local storage and vector search
- RAG over personal study material
- Python-based automation for useful local actions
- modular architecture for future tools and integrations
The goal is not to build a new foundation model.
The goal is to build a reliable personal AI layer that can understand context, route tasks, remember useful information, and assist with daily technical workflows.
J.A.R.V.I.S. is designed to:
- receive voice commands through an Alexa-based or speech interface
- route intents through a Python backend
- use Ollama for local LLM responses
- call Claude or external models as fallback for complex reasoning
- store and retrieve persistent memory
- search study documents through a local RAG pipeline
- generate contextual responses based on user state
- run lightweight PC automation actions
- expose a dashboard or frontend for monitoring and interaction
Voice Input
Alexa / Speech Interface
↓
Python Backend
Intent routing · State management · Response building
↓
Reasoning Layer
Ollama local LLM · Claude fallback · Assistant logic
↓
Tools Layer
PC automation · File actions · Study assistant · Utility functions
↓
Memory Layer
ChromaDB · Local JSON memory · RAG over documents
J.A.R.V.I.S. uses local models through Ollama whenever possible.
This keeps the assistant:
- low-cost
- private by default
- available without depending entirely on cloud APIs
- suitable for experimentation with local AI workflows
For more complex reasoning tasks, the assistant can route requests to external LLM providers such as Claude.
The idea is not to replace local reasoning, but to use cloud models selectively when they add real value.
The assistant can store and retrieve useful context over time.
Memory can include:
- user preferences
- study context
- repeated workflows
- useful notes
- structured information for future responses
J.A.R.V.I.S. can be extended with retrieval over local documents, allowing it to answer questions based on personal study material or project files.
This makes the assistant more useful than a generic chatbot because it can work with actual context.
The project includes the foundation for local automation actions such as:
- opening files
- managing folders
- triggering scripts
- interacting with local workflows
- assisting with repetitive computer tasks
This project is a practical experiment in building personal AI systems that are:
- local-first
- modular
- privacy-conscious
- low-cost
- extensible
- useful in real daily workflows
Instead of relying on a single chatbot interface, J.A.R.V.I.S. combines multiple components into a small personal operating layer.
- Python
- Flask / FastAPI-style backend architecture
- Ollama
- Local LLMs
- Claude fallback
- ChromaDB
- JSON-based local memory
- RAG pipelines
- Voice interface experiments
- PC automation scripts
The long-term goal is to evolve J.A.R.V.I.S. into a practical personal AI operating layer.
Future directions include:
- better intent routing
- stronger memory management
- safer tool execution
- improved voice interaction
- dashboard interface
- task scheduling
- project-aware assistance
- deeper RAG over local files
- modular plugin architecture
This is an experimental personal AI project.
It is built as a learning and portfolio project to explore:
- local-first AI assistants
- LLM routing
- voice-based interaction
- persistent memory
- retrieval-augmented generation
- AI-assisted productivity
- personal automation
The project is under active development.
This repository is meant to show practical experimentation with AI assistant architecture, not just prompt design.
The focus is on building a system that connects:
User context → Reasoning → Memory → Tools → Useful action
Built by Marco Antonio Carmine Abate.
AI & Data Science student focused on agentic workflows, LLM evaluation, local-first assistants, and AI-assisted software development.
- GitHub: TacoengineerIT
- LinkedIn: Marco Antonio Carmine Abate