Vasudev AI – Hinglish JARVIS for Everyone

Vasudev AI Light UI
Vasudev AI Dark UI

🚀 Inspiration

As a developer and student from India, I always wanted a truly personal AI assistant like JARVIS — but one that respects user privacy, works offline, and understands our Hinglish way of speaking.
With limited internet access and expensive cloud-based tools, I saw the need for a desi-smart solution.
That’s how Vasudev AI was born — built to run powerful AI agents on-device, speak Hinglish fluently, and function fully offline.

🤖 What it does

Vasudev AI is a personal offline AI assistant designed for Indian users.
It runs entirely on your own device, using powerful open-source AI models.

🧠 Key Capabilities:

Smart conversational assistant powered by LLaMA 3.2 GGUF model
Background remover AI using Python and OpenCV (coming soon)
Voice-based chat interface (coming soon)
Reads and processes Excel files for productivity tasks (coming soon)
Hinglish support for regional language accessibility
Fully private: No cloud, no tracking, no data sharing (for myself)

🛠️ How we built it

We built Vasudev AI using open-source tools and a focus on edge-device compatibility:

Backend: Python + Flask handles core logic and APIs
Frontend: Simple HTML/CSS interface with chat support
AI Runtime: LM Studio runs LLaMA 3.2 3B GGUF model locally
Background Remover: Built using OpenCV and Pillow (PIL)
Server Logic: When Flask server is inactive, the domain shows a fallback static HTML page hosted via GitHub
Tunnel Access: Cloudflare Tunnel allows secure remote access via browser

⚔️ Challenges we ran into

Model Optimization: Running large LLMs on a low-end Windows laptop required model quantization and performance tuning
Dynamic Hosting: Creating an auto-switch DNS rule system between Flask and static GitHub page
Hinglish NLP: Building a chat system that could understand code-mixed Hinglish using prompt engineering
Privacy-by-Design: Ensuring 100% on-device AI functionality without cloud APIs, while maintaining responsiveness

🏆 Accomplishments that we're proud of

🧠 Built a fully offline, intelligent AI assistant for Indian users
🧩 Integrated multiple AI agents (chat, background remover, Excel reader) into one offline toolkit
🌐 Created a smart fallback hosting system using GitHub + Cloudflare DNS
💻 Got everything running smoothly on a basic Windows laptop

📚 What we learned

🧪 Mastered quantized model serving with GGUF and LM Studio
🔧 Improved skills in Flask, local AI deployment, and edge system design
🌍 Learned to handle Hinglish NLP, memory optimization, and UX on constrained devices
🔐 Gained experience in network tunneling and dynamic domain management

🔮 What's next for Vasudev AI

The journey doesn’t stop here. Upcoming milestones include:

🗂️ Add tools like PDF summarizer, AI calling agent, and text-to-speech
🧑‍🏫 Train a custom Hinglish dataset for more natural, regional conversations
🧠 Implement on-device memory + contextual awareness
📱 Build a mobile app version for Android devices
🌐 Release open-source modules for the community to build upon

Vasudev AI is not just another AI assistant —
It’s a private, local-first, and intelligent desi alternative to cloud-based assistants.
Built in Bharat. Made for Bharat. 🇮🇳

Built With

cloudflare
css
flask
html
llama
lmstudio
python

Submitted to

FusionHacks 1

Created by

Created by
Surya Pratap Singh is the creator and driving force behind Vasudev AI. As a developer, tech enthusiast, and student, Surya envisioned a truly private, offline AI assistant tailored for the unique needs of Indian users.

My Contribution
Project Concept & Vision: I conceptualized Vasudev AI as a solution that would bridge the gap between cutting-edge AI technology and the regional needs of users in India. My vision was to build an AI assistant that respects user privacy, works offline, and understands Hinglish, which is the language blend commonly spoken in India.

End-to-End Development: I led the development process, from the initial planning stages to the final implementation:

AI Model Integration: I integrated the LLaMA 3.2 GGUF model for local conversational abilities, using LM Studio for model hosting and execution.

Backend Development: I implemented the backend logic using Python and Flask to serve the API endpoints and manage communication between the frontend and the AI engine.

Image Processing Module: I developed the background remover feature using OpenCV and Pillow (PIL) to handle image editing tasks.

Frontend Development: I built a simple, user-friendly interface using HTML/CSS for chat functionality and interaction with the assistant.

Privacy and Offline Capabilities: I ensured that all features of Vasudev AI work entirely offline, prioritizing privacy and security by eliminating the need for cloud-based services.

Server Management: I set up the Cloudflare Tunnel to allow access to the local server from any device and managed a fallback system using GitHub Pages for static page hosting during downtime.

Future Development & Strategy: I defined the roadmap for Vasudev AI, including plans to add more features like voice interaction, a mobile app version, and open-source releases of core modules.

My Goals
I aim to continue enhancing Vasudev AI with more advanced tools, further improve its multilingual capabilities, and ultimately offer it as a robust, open-source tool for users everywhere.

Surya Pratap Singh

Updates

Surya Pratap Singh started this project — May 21, 2025 02:26 AM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.