DataVision

Inspiration

We wanted to give people the ability to easily discover interesting relationships and obtain actionable insights from any data they have, regardless of their data science prior knowledge/background.

What it does

The user uploads a CSV file and the agent cleans the data, comes up with potentially interesting relationships that might exist in the data, performs hypothesis testing to see if those relationships are actually there, creates visualizations to see these relationships, and provides a plain-english summary of everything that it found. After this, the user can ask follow ups about the analysis related to anything in the data and the agent will answer.

How we built it

For our frontend, we used Next.js with TailwindCSS, building a real-time drag-and-drop interface and a dynamically rendering interface for the data analysis and user follow ups. For the backend we used Python + Flask to power a microservice that cleans data, runs tests, and streams p-values and visuals, using Google Gemini with LangChain and LangGraph for our agentic workflow.

Challenges we ran into

Some challenges we had were syncing frontend-backend data streams and coordinating inputs and outputs for our agentic workflow.

Accomplishments that we're proud of

We built an intelligent agentic data exploration tool that can find interesting relationships in any data with a very clean UI.

What we learned

We deepened our understanding of real-time data streaming, backend/frontend orchestration, and agentic AI systems.

What's next for DataVision

We aim to improve our system by creating even more detailed and relevant visuals, adding more useful information about the analysis procedure conducted, and making our data analysis agent more intelligent and capable overall.

Built With

• azure
• gemini
• langchain
• langgraph
• next.js
• python
• react.js
• tailwind
• typescript