Roster.AI

RosterAI leverages AI/LLM and Web Scraping technology to provide real-time advanced Fantasy Football insights and decision-making

Comment

Inspiration

Each of our team members entered HackGT as solo participants, but we quickly found common connection in our love for fantasy sports. As casual fans and students with not a lot of time on their hands, we wanted to build an app to leverage LLM technologies to bring the fantasy experience to all.

What it does

First, the user imports their NFL fantasy roster as a csv (format: playername, POS, Slot, Team, Proj, Opp), which is then processed by our application and saved in local storage. At any point, the user can access their fantasy rosters. From there, our two primarxy features are our real-time news sentiment analysis and players insight page. For the former, the app reads your current rosters players and searches online for articles relating to that player and their fantasy performance. Then our LLM analyzes the article content and displays how that information impact s the player's performance in the future. Then, our player insights page combines the information created in the news analysis and provides per-player insights and outlooks, as well with recommendations for roster moves.

How we built it

The front-end and UI/UX was built using TypeScript/React-native and Expo for iPhone. Additionally, for the web scraping we used Selenium and BeautifulSoup. For the LLM portion, we integrated with Google Gemini's API. For middle-end API connections, we used FastAPI to connect our LLM and Web scraping pipelines with our front-end display.

Challenges we ran into

Implementing web scraping with a Python script for the first time as a group, while building out the UI for the React Native application, all while also combining both stacks using FastAPI. Researching different APIs and trying to find different methods of obtaining data from third party resources. How to use Git without messing eachothers code and workflows. Figuring out how to integrate with LLM AI for the first time. Difficulty with prompt engineering and Gemini, figuring out how to get desired outputs in a reasonable amount of time.

Accomplishments that we're proud of

We are particularly proud of how we were able to create a robust UI that simplified the entire process of competing in a fantasy league as well as enhance the causal user’s experience. Also, we were proud of the integrations of a web scraper we designed to find news related to a player while having Gemini summarize the article for the casual reader to read. We are proud that we are able to help casual fantasy players even the playing field with analytics and assistance.

What we learned

We learned how to use Web scraping/Data Mining and Text Processing to efficiently find, extract, and organize domain-specific knowledge from large-scale online content. We learned how to integrate API's (Gemini) to create desired outputs given information from web scraping. We learned how to use GitHub as a group (creating repositories, pushing, pulling, using branches, merging, etc) and how to work with React Native in VSCode.

What's next for Roster.AI

Potential Ideas for Improvement: Integration with popular Fantasy websites (PrizePicks, ESPN Fantasy, Sleeper) Increase Web scraping scope Further develop LLM-powered insights Include data from NFL matches using NFL's API Include Injury Status & Reports in Decision-Making

Share this project:

Updates