Niam LeStourgeon

Agents / Analytics / AI

Fantasy Football Agent

The AI that researches with your brain, so that you don't have to.

The Numbers

218
Model Calls
119
Decisions Served
48
Decisions Graded

See it in action: a 56-second walkthrough of how Fantasy Agent researches, decides and grades its own calls, built and shipped end to end.

Visit the live site →

The Problem

I love football, but to be good at fantasy football, you need to spend hours watching the games, doing research, really stay on top of the news if you want to be competitive. Being a student, working a job, spending time with family makes this intense research difficult and in turn takes away your chances to win your fantasy league.

How It Works

Your Philosophy

You write your philosophy, what you would do in every situation. This is stored to your account, and it is referenced before the agent makes any decisions.

Your Schedule

You choose when you want the emails in your inbox. You can have different engines run on different days, such as just the lineup engine on Sunday mornings when you need it most!

RAG Pipeline Activates

An AWS Lambda function kicks off the process, retrieving all the data from news sources, sports APIs and even the weather!

AI Takes Over

The news is passed into Jev, to decide what articles have fantasy relevance. It excludes any article that is not directly relevant, like a reflection piece. Then Claude takes everything in and analyzes all the data, and makes the recommendations based on your set philosophy.

Straight to Your Inbox

You get your research in your inbox, ready to judge and implement in time for the games.

Better Than Yesterday

All of the AI's calls are graded on how well they worked out. If it makes a bad call, it learns from it and improves for your next week.

Key Decisions

Jev Before Claude

Putting Jev before Claude was huge. Jev came out for the first time right as I started this project, so I was uncertain about how well it would work. Jev is a decision model (different from LLMs) that doesn't return text, it returns a decision and a confidence score.

In this project it looks at football articles and determines if an article is relevant to fantasy football or not. This is crucial, because all of the words of the article get passed into Claude, which charges for input "tokens" (a few letters, the measurement of AI computing power) Jev charges a fraction of the cost for input tokens, so using it to filter out the news articles saves me a ton week to week.

Only Using AI Where It Makes Sense

There are a lot of "ChatGPT wrappers" out there that are just a pretty ui around ChatGPT or another such chatbot, which are the ones we're all tired of. The real strength of this project is in two parts:

  • The research element. Doing all that research to who the best available player is takes a ton of time. This cuts all that out.
  • Your philosophy. A generic ai doesn't understand how you like to play. Your philosophy that you make guides every decision, which checks to ensure the AI doesn't deviate.

Both of these make this the perfect use for AI for Fantasy Football. No overuse of AI, just using it where it makes sense. AI integrates into the system to play to its strengths, it doesn't take over the whole system.

What's Next

The AI currently doesn't help at all for draft, which is arguably the most important part of a good fantasy football team.
What's next for me is the Tick Forward project and continuing in my job search.

Tech Stack

Python FastAPI PostgreSQL AWS Lambda/EventBridge/SQS AWS CDK Claude Code and API Jev Stripe