Case Studies

Our client approaches us to develop a smart assistant for sport bids analysis and predictions based on ChatGPT. The core functionality revolved around processing large amounts of historical data to provide accurate answers to users' questions.

One of the main challenges of this project was the sports data: typically, large LLMs analyze text data, while sports statistics are presented in a form of spreadsheets with numeric parameters.

We have created a flexible and reliable smart assistant capable of answering complex user questions about sports statistics, as well as perform predictive analytics and offer bidding data.

Our solution features an intuitive chatbot-like user interface, an LLM, an API gateway, a data processing module, a vectorization module, a PostgreSQL vector database and a connection to an external sports statistics provider.

Complex Data Analytics

We have integrated OpenAI's GPT to process user requests and perform complex analytics: frequency questions, questions requiring calculations, like mean values and probability estimations.

Our team has fine-tuned GPT to generate personalized responses, offer users accurate sports statistics, predictive analytics and betting information.

The AI model improves with use as it trains on the conversations it has had.

Chabot UI

We have designed the smart assistant UI to resemble the most popular chatbots to make user interaction with the system intuitive.

The system is open source to facilitate simple chatbot integration into our client’s systems and website, as well as provide flexible customization as the project evolves.

Statistical Data Management

The application relies on sports data provided through reliable APIs from renowned sports data providers.

Data providers often have a limit on the number of requests per minute, so it was not possible to establish operational interaction via the API. Given the restrictions, we have created a database which updates with new data as it becomes available.

The LLM communicates directly with an internal database, avoiding data providers’ restrictions. This approach makes the system easily scalable and flexible, accounting for the potential user growth in the future.

Data Preprocessing

Statistical data requires preliminary processing, like data cleanup and organization to prepare the data for the LLM to effectively process it. Data preprocessing ensures accurate data analysis and answers for users.

In addition, we have trained GPT to generate SQL queries to obtain numerical data from the database.

Users who place sports bets on a regular basis with our client have received a powerful tool for accessing sports statistics and analytics in chat mode with an AI assistant.