The Point: Where every fan is a play caller

Inspiration

There's something to be said about the experience of watching a sporting event that comes down to the wire, the excitement and anxiety that radiates from both teams is something you can't get anywhere else. Picture this: your favorite team has had a historic season and made it to league finals, and the game is neck and neck. Double OT, tied game, your home team has one shot at victory, everything riding on the next shot, but you're not stressed, because you knew the game was over as soon as the star player got the ball. They take the buzzer beating shot, and your team are officially champions . Moments like these are the lifeblood of sports, they're what every viewer is itching for everytime they turn on the TV and tune in, and they are the core focus of The Point.

Our platform allows users to test their intuition in the most crucial moments of the game as well as provide insights as to which potential outcomes would swing the game odds the most, and share these moments with each other.

What it does

The Point transforms real-time football data into an interactive, predictive experience:

We start with a live feed for any NFL game.
As each play unfolds, an AI reasoning agent evaluates live context (score, time remaining, field position, down) and enumerates possible outcomes to generate possible scenarios as to how the game will unfold next.
A statistical model uses context to compute how each outcome shifts a team’s win probability.
Users choose from those outcomes and win probabilities, making a prediction and betting their points on what they believe will happen.
When the play resolves, users earn points based on prediction accuracy and how big the win-probability swing was.
A real-time leaderboard keeps the competition alive, turning every play into a strategic, communal moment.

How we built it

User Platform

Our frontend is build off of Next.js and TypeScript for ease and speed of development as well as compatability with our other technologies. By using the Next.js API router system, we were easily able to create endpoints that are accessible through http and would help us structure our data pipeline between the data API, the scenario agent, the statistic model, and the validation system.

Sports Data Feed

As a source for our scenario generation agent, we decided to use Sportsdata.io because their APIs work for both live and final games, allowing us to create a demo and allow for easy implementation into actual live games in the future. The data provided by the API provides full context on the latest game events and states in near real-time, allowing our scenario generation to take place at a reasonable timing.

Scenario Generation and Probability Shift

We didn't have time to train an entire model for generating possible outcomes, so we went with a simpler approach of using a singular game state provided by the Sportsdata API, reasoning what state the game was in using the context, generating possible outcomes of the the play, and use our agent to create new proposed values of what the game state would be in that scenario. We then feed the potential outcomes and the original game state to a simple linear regression model that computes the probability the possession team has of winning the game and subtracts the original from it to calculate the shift in win probability.

Scenario Validation

This was an interesting problem, since finding a correlation between the generated game states and the actual outcome to draw a conclusion on if the selected scenario occurred was difficult to do consistently. Our solution was to look at the description of the scenario selected and use LLM reasoning to compare it to the description provided by the sports data feed of the actual play to validate a users selection.

Challenges we ran into

Mocking real-time gameplay: Since no live games were available during the hackathon, we had to creatively simulate a live feed using historical play-by-play data while ensuring that updates still felt dynamic and authentic for the demo.
Integrating components quickly: It was some of our first experiences with new technologies, and connecting the frontend, API, and models under tight time constraints was a hard but rewarding learning curve.
Scouring the internet for publicly available and usable NFL data with play-by-plays was somewhat time-consuming.
Creating/Using a statistics model that we could create on a timely manner, use in general scenarios with the information available from Sportsdata, and execute quickly to provide real time analysis.
Validation: Consistently finding a correlation between the generated scenario selected by the user and the actual game event.

Accomplishments we're proud of

We're really proud of building a functional prototype that ties together real data, AI reasoning, predictive modeling, and gamified engagement in under 36 hours despite the challenges and differing levels of experience we had. We also created a realistic mock “live” football experience using past games to showcase the platform’s potential. I believe all of us collaborated really well with our individual strengths, and helped each other out when we were stuck to deliver something awesome.

What we learned

The power of merging reasoning AI with data-driven modeling to make sports analytics engaging and understandable.
Gained hands-on experience with sports data pipelines — cleaning, structuring, and extracting insights from complex play-by-play datasets.
Thinking from user perspectives and focusing on their painpoints and interests
Learned how to build AI agents that reason through dynamic, context-rich scenarios.
Strengthened our skills in Next.js for responsive, interactive web interfaces.
How to work with what you're given and come up with unique solutions to challenges
That clear task division, fast iteration, and tight integration are essential during hackathons.

What's next for The Point

What's exciting about The Point is that there is so much room for growth and we had so many ideas that we simply didn't have time to implement. As far as our product's core engine, we could improve almost every component. By paying for a more premium sports data provider, we can receive more contextual data faster, allowing our generation and reasoning models to produce better results, We could train a custom model for creating the possible scenarios rather than using simple reasoning from a game state, and our statistics model could operate with more computing and improved from a simple linear regression to more accurately represent the probability shifts with more context provided.

As far as our platform and user experience, as opposed to the demo, The Point is meant to be a live game interactive experience, and would be built off of this idea. As a user watches their sports feed from their providers through our platform, they'll gain points that they can use to bank on their next prediction. Predictions will present themselves to the viewer at random throughout the game for a user to use points on, with later predictions increasing point payouts. These points you accumulate through watching games and interacting will help populate leaderboards and other data so the user can compare their intuition to other people and post about it on social media. Eventually, we could fully lean into the social platform and add more viewer interactions and even other sports.