Database Architecture
Our database is designed by professionals to ensure quality & edge to include and surpass NBA teams’ data sources. We work directly with NBA-calibre analytical professionals, tech executives, coaching analysts, and players to ensure on the quality of the data.
Data Sources
Pro Advanced Statistics via APIs
In TracyPro, you get full access to her Professional-Grade Advanced Statistical Database that is not only built out from NBA-calibre API data sources like SecondSpectrum, CleaningTheGlass, Sportsdata.io, SportRadar, and a few other currently undisclosed sources.
All of this data acts as her knowledge base, where she can refer to specific previously publicly unavailable data and allow all users to gain insights that are digested and analyzed for you by Tracy.
Real-Time Twitter & Article News Sentiment Data
Tracy is constantly upgrading her knowledge based on the latest injury, trade, and news information on the market. This includes multiple news sources such as the latest sports journals and also beat writers and major NBA insiders that provide Tracy with the timeliness she needs to comment on the most recent trades.
Proprietary performance metrics & statistics
We have a team of NBA analytical experts that act as advisory to support Tracy's analytical capabilities as well as helped us develop multiple custom performance metrics that allow Tracy to effectively summarize & analyze performances of teams, players, and more.
Currently this information is not public however we will gradually release more information on our custom metrics in the future.
Web indexed information
As part of a backup to round out her basketball knowledge, Tracy pulls from a massive database of web-index information to provide her the additional context that will allow her to answer more qualitative questions.
Architecture & Storage
Our data storage is divided into structured and unstructured parts. Structured data is stored using PostgreSQL, mainly covering various NBA teams and player game statistics. We can transform the mathematical components of user queries into database queries, leveraging the powerful computational capabilities of the database. This approach avoids hallucinations common in large models and provides statistical and large-scale data processing capabilities that LLMs may not possess.
Unstructured data is distributed across Milvus DB and AWS Bedrock Knowledge Base. Milvus DB primarily stores domain knowledge related to the NBA, while the AWS Bedrock Knowledge Base contains NBA news and articles collected from Twitter, blogs, and sports websites like ESPN. Tracy extracts relevant knowledge from Milvus to analyze user questions and combines this with data retrieved from PostgreSQL and the Bedrock Knowledge Base to construct the informational context. Finally, a complete answer is generated using an LLM.
Last updated