Technologies
TypeScript, Next.js, Payload CMS, Python, FastAPI, PostgreSQL
AI: OpenAI, Multi-Agent Systems, Intent Classification, Dynamic Toolin
duration
4 Months
customer
The client is a major player in the horse racing industry. The goal was to create a central hub for all horse racing data and an intelligent AI companion to provide deep insights for owners, trainers, and enthusiasts, dramatically improving decision-making and access to information.
Background and problem
The client needed to aggregate data from numerous disconnected sources—including live racing APIs, weather data, scraped blog posts, and internal databases—about horses, their health, training, and global race results. Beyond just a database, they required an AI-powered solution that could understand complex natural language questions and provide information-aware answers on any racing subject, from a horse's training quality to future racing schedules.
solution
We delivered a comprehensive platform consisting of a custom CMS, a public-facing website, and a sophisticated multi-agent AI companion.
The AI system intelligently processes user inquiries by first classifying them into topics, intents, and entities. It then generates a dynamic action plan, selecting from a suite of tools to retrieve data from vector stores, scraped websites, internal knowledge bases, racing APIs, and weather APIs. The solution features dynamic, RBAC-based instruction building, allowing admins to customize AI behavior and available tools for different user groups.
The main challenge was creating a reliable, multi-step AI that could reason over diverse, real-time data sources to generate accurate and context-aware responses. We successfully implemented a robust architecture with full test coverage for both AI and functional parts, ensuring high reliability. The platform now successfully manages and syncs data for thousands of horses and races from all over the world.

