Data Analyst

This project I have made involves using AI Agents powered by a Groq API to achieve great analysis in data that is submitted through a data file. Visit it: https://data-processing-ai-agents.vercel.app/

Comment

Inspiration

Despite using computers as a computation mechanism, it never occured to me to try and utilize the recent computational developments to solve bigger issues, primarily data related. My dad uses AI all the time [sadly he chooses copilot, the corst choice he could make!] to get the analysis of the data he is looking at. This project isn't just for him, however. This project taught me how to use agents outside of the AI apps that I use such as OpenSwarm and Claude Code, helping me understand better how these apps work and developing my skills.

What it does

This AI Agent project takes in data from the frontend [a react vite site hosted on (vercel here)[https://data-processing-ai-agents.vercel.app/]] and pushes it to the agent-run backend on hugging face that processes the request by breaking down the context into what the user wants and what the user gave context about. Then it goes to the agent that creates the custom prompt for the data analyst agent, which then the data analyst looks at the data and then analyzes the data by looking at outliers and summarizing the data points in regards to the context. Then the output agent formats the response from the data analyst into how the data should be represented for the use case provided and then rates the response given by the data analyst compared to what the user requested, allowing the user to see if the agent believes the question was properly answered based on their prompt.

How we built it

I used react vite JS for the frontend, a REST API connecting the frontend to the backend through the use of async functions, and CrewAI as the backend agent manager and creator. The frontend is hosted on vercel and the backend is hosted on HuggingFace. I used any LLM API that I could get my hands on for free.

Challenges we ran into

There were many challenges in this process including trying to run the frontend properly, LLM models kept returning errors despite me using the OpenRouter API properly, it was very difficult to make work. Also the testing phase took a while to load since the AI Agents thought deeply which is good however it made my job a bit long. However all of those issues seem so minor compared to the API keys that would max out so quickly while testing, I remember wondering if there was any time to pivot and create a different project.

Accomplishments that we're proud of

This is a great project considering I'm currently studying for physics and precalculus finals for school as of submitting this project, so I hope that my studying pays off as much as I know this project will. I am also proud of the fact that this project didn't turn out a complete failure, it was quite nice to build and even use for processing my own data.

What we learned

I learned about CrewAI and I realized how much of a game changer it really is for me to use, it creates agents and manages the agentic workflow flawlessly except for my own errors and poor prompting that get thrown in. I've started to implement CrewAI in my own repos more and especially repos that include AI Agents being used.

What's next for Data Analyst

I have been working on making this Data Analyst project used in all of my major GitHub repos that have the ability to utilize AI in this way, ranging from clustering to explaining growth. I plan to use this as the backend in my next competition PhysTech 2026 as the processing force however I am not taking everything and copy-pasting it.

Built With

Share this project:

Updates