Home
Projects
The Lab
Photo Lab
Contact
Back to Projects
Case Study

Project JARVIS

Project JARVIS

Problem

Standard large language models lack deep, personal context. While they are great at general knowledge, they cannot easily integrate with personal projects, local files, codebase structures, or calendar events without significant manual input or high latency.

Insight

By leveraging local inference models connected to a personal database using Retrieval-Augmented Generation (RAG), it is possible to build a unified AI operating system that deeply understands a user's specific workflow and context without compromising privacy.

Process

I began by evaluating various local LLMs, focusing on recent models like Llama 3 for their balance of speed and reasoning capabilities. The architecture was designed to automatically ingest local markdown files, codebase structures, and daily notes, embedding them into a local vector database for rapid retrieval.

Solution

Project JARVIS is an ongoing exploration into a true personal AI system. It features a seamless interface for querying personal data and active codebases. The unified memory architecture ensures that context delivery latency is minimized, allowing for fluid and highly relevant interactions directly within the development environment.

Outcome

While still in active development, the system already successfully indexes the local file system and accurately answers complex queries regarding ongoing projects. This drastically reduces the time spent searching for notes, debugging old code, or trying to rebuild context.

Lessons