They Bet, They Lose, They Learn: The Agents That Rewrite Their Own Playbook

An agent learns to predict World Cup 2026 game outcomes by using and iteratively improving its own playbook based on match results.

Gemini MCP server Tavily football-api LLM

Overview

An Agentic Self-Harness that makes predictions about the WC 2026 game results using a playbook, analyzes match outcomes and improves the playbook that it uses for the next prediction round.

Links

https://cup-clash-international.lovable.app/
Cup Clash is a free, points-based World Cup prediction game.

Tech stack

Gemini

Google's natively multimodal AI model: understands and operates across text, code, audio, image, and video.

Gemini is Google's most capable and general AI model, engineered from the ground up to be natively multimodal: it seamlessly understands and combines information across text, code, audio, image, and video inputs. The technology is optimized for flexibility, running efficiently on everything from data centers to mobile devices. It is deployed in three key sizes: Ultra (for highly complex tasks), Pro (for broad scaling), and Nano (for efficient on-device tasks). Developers access this power via the Gemini API to build next-generation applications.

https://deepmind.google/technologies/gemini/

View projects
MCP server

The MCP Server implements the Model Context Protocol: a standardized interface connecting large language models (LLMs) to external tools and real-time data.

The Model Context Protocol (MCP) Server acts as the critical bridge—a 'USB-C port for AI'—that extends an LLM's capabilities beyond its training data. It uses a standardized protocol to expose real-world functions and resources to AI agents, enabling direct interaction with external systems. Specifically, the server advertises tools (like running code or searching the web) and resources (like local files or databases) via a consistent JSON-RPC interface. This allows AI clients, such as Claude Desktop or Cursor, to perform complex, contextual tasks: for example, automating GitHub workflows or querying a MongoDB instance on command.

https://mcpservers.org

View projects
Tavily

Tavily is the AI-optimized search API, purpose-built to deliver real-time, structured web data directly to your Large Language Models (LLMs) and RAG pipelines.

Tavily functions as the essential web access layer for AI agents, providing scalable Search, Extract, and Crawl APIs designed specifically for LLM and RAG workflows. It bypasses traditional search engine limitations (raw HTML, unstructured snippets) by reviewing multiple sources, intelligently synthesizing content, and delivering clean, structured data with citations. This process ensures your AI applications (chatbots, agents) receive accurate, real-time, and production-ready information, significantly reducing model hallucinations. Integration is fast: utilize the Python or Node.js SDKs, and start testing immediately with the generous free plan.

https://tavily.com

View projects
football-api

A RESTful API delivering real-time football data, live scores, and statistics for over 1,000 global competitions.

API-Football provides developers with a highly reliable stream of global football (soccer) data through a clean RESTful architecture. The service covers more than 1,000 leagues and cups (including the Premier League, La Liga, and the UEFA Champions League) with live event updates refreshed every 15 seconds. Developers can easily query endpoints for real-time livescores, standings, team lineups, player statistics, pre-match odds, and historical data spanning over 15 years. With ready-to-use widgets and SDK support across major languages like Python, Node.js, and PHP, it simplifies building sports apps, fantasy platforms, and live scoreboards.

https://www.api-football.com

View projects
LLM

Large Language Models (LLMs) are deep learning models, built on the Transformer architecture, that process and generate human-quality text and code at scale.

LLMs are a class of foundation models: massive, pre-trained neural networks (often with billions to trillions of parameters) that leverage the self-attention mechanism of the Transformer architecture (introduced in 2017) to predict the next token in a sequence. Trained on vast datasets (e.g., Common Crawl's 50 billion+ web pages), these models—like GPT-4, Gemini, and Claude—acquire predictive power over syntax and semantics. They function as general-purpose sequence models, enabling critical applications such as complex content generation, language translation, and automated code completion (e.g., GitHub Copilot). Their core value: generalizing across diverse tasks with minimal task-specific fine-tuning.

https://en.wikipedia.org/wiki/Large_language_model

View projects

They Bet, They Lose, They Learn: The Agents That Rewrite Their Own Playbook

Related talks