Configuring a workspace
System Logging
Chatting
Document Upload to system and saving to a workspace

Trek AI

Inspiration

We initially aimed to create a browser automation tool, but pivoted when we realized the limitations of open-source models in controlling browser navigation. Inspired by tools like AnythingLLM, we developed Trek AI - a private, easily deployable enterprise data interaction tool using NVIDIA Workbench.

Demo Video

Trek AI Demo Video

What it does

Trek AI is an AI-powered platform built on NVIDIA AI Workbench that allows enterprises to interact with their data using natural language. It offers:

Document processing and analysis
Natural language querying of company information
Collaborative data-driven tasks
Complete data privacy and security
Configurable AI agents for various tasks
Custom tool creation for agents
Data visualization and graph generation

How we built it

We used NVIDIA AI Workbench to implement the following technologies:

Astro with SSR for backend and frontend
Lucia Auth for authentication
SQLite for structured data
LanceDB as a vector database
Tailwind CSS for styling
Vanilla UI for components

Challenges we ran into

Integrating local AI models with NVIDIA AI Workbench
NVIDIA Workbench proxy pathing issues, requiring a custom Vite plugin (astroPath) for Astro
Balancing feature development with user-friendly design
Time management and rapid pivoting of ideas

Accomplishments that we're proud of

Developed a full-fledged enterprise AI tool in just 2 weeks
Implemented a comprehensive system including:
- User authentication
- Workspace management
- Configurable AI agents
- LLM integration and configuration
- Custom tool creation for agents
- Data visualization with graph generation
Created a secure, private data interaction system
Developed an intuitive natural language interface for data querying and agent interaction
Successfully integrated with NVIDIA AI Workbench, leveraging its powerful capabilities
Overcame technical challenges like NVIDIA Workbench proxy pathing issues

What we learned

Helpfulness of creating prototypes to show to teammates to help guide project if we were to do it again I would have focused more on this.
Time management and prioritization in a high-pressure, short-timeline project.
The potential of AI agents and LLMs in transforming enterprise data interaction
Overcoming technical hurdles

What's next for Trek AI

Add functionality for users to create custom tools with their own code and call their own APIs
Implement a system to fine-tune AI models within Trek
Finish implementing local AI processing with NIM
Integrate data sources imports like Jira, GitHub, Notion, and website content
Integrate the browser extension from initial project for enterprise web automations
Expand supported document types
More collaboration features and things like inviting people to workspaces and what not.
Developer API to call Trek AI anywhere.

Trek AI aims to make advanced AI capabilities accessible and secure for businesses of all sizes, revolutionizing enterprise data interaction. By providing a powerful, flexible platform for AI-driven data analysis and interaction, we're enabling companies to unlock new insights and efficiencies in their operations.