JRVS

• GIF

• 

Inspiration

Inspired by Iron Man’s iconic AI assistant JARVIS, we wanted to bring that same experience into real life — a personal AI that learns from you, understands your habits, and helps you manage everything effortlessly. Our goal is to create a conversational assistant that reduces “mental overhead” and lets you focus on what truly matters.

What it does

JRVS (Jarvis) is a continuously learning AI assistant that grows smarter over time through chat interactions. Currently, it runs as a local PC-based software that stores, recalls, and uses your inputs to automate tasks like scheduling, searching, and summarizing.

For example:

Add a calendar event by simply typing: “Schedule a meeting tomorrow at 1 PM.”

Summarize and learn from any web page by typing: /scrape

Instantly switch between AI models depending on the task

Search your personal knowledge base for relevant information

It’s your personal AI that learns like a snowball — getting more capable the more you use it.

How we built it

JRVS is built with Python and powered by a modular, performance-optimized architecture:

🧠 Ollama LLMs — Local large language models for fast, private AI interactions

🔍 RAG (Retrieval-Augmented Generation) — FAISS + BERT embeddings for semantic context retrieval

🌐 BeautifulSoup — Automated web scraping and content indexing

💾 SQLite — Persistent memory for chat history and document storage

🎨 Rich TUI — Stylish terminal UI with switchable themes (Matrix, Cyberpunk, Minimal)

⚡ Async + caching — Lightweight, fast, and efficient

Directory structure:

core/ → Database and performance optimization
rag/ → Vector search and context retrieval
llm/ → Ollama model integration
cli/ → Command-line UI and command handling
scraper/ → Web scraping functions
data/ → Persistent database and FAISS indexes

💥 Challenges We Ran Into • Integrating multiple APIs was trickier than expected — managing authentication tokens and handling rate limits required careful debugging. • We faced time constraints balancing backend setup and UI polish, so we had to prioritize core features over extras. • Some third-party libraries didn’t work as documented, forcing us to find alternative solutions on short notice. • Collaborating remotely made version control and merge conflicts a challenge until we standardized our Git workflow. • We had to learn new frameworks (e.g., React hooks, Flask routing, Firebase) on the fly to implement our ideas.

🏆 Accomplishments That We’re Proud Of • Built a fully functional prototype in under 36 hours with working authentication, database integration, and responsive design. • Designed a clean, intuitive interface that users immediately understood and enjoyed testing. • Implemented a feature (e.g., live tracking, AI-based recommendation, or automation) that exceeded our initial goals. • Overcame technical and time barriers by working together and leveraging each member’s strengths. • Deployed a working demo that’s accessible online and actually useful beyond the hackathon.

What we learned

True intelligence comes not from the LLM alone, but from context augmentation through RAG.

Even small UX improvements can drastically change how natural and enjoyable AI interaction feels.

Integrating scraping, vector search, and local models is a major step toward offline AI ecosystems.

🚀 What’s Next for JRVS • Expand our project by adding new features like analytics dashboards, user customization, or mobile support. • Optimize our backend for scalability and security before launching a public beta. • Gather user feedback to refine our UI/UX and prioritize features for v2.0. • Explore partnerships or open-source the project so others can contribute. • Continue developing JRVS as a long-term product, potentially turning it into a startup or research project.

Built With

• github
• next.js
• ollama
• python
• typescript