The session is titled Serving Delta via API. It explores the rising popularity of Delta Lake and various methods of serving delta data through APIs.

It begins by dissecting the structure and concept of Delta Lake, highlighting its growing adoption and versatility in data processing. After that we examine different Python libraries like Polars and Pandas for reading Delta data and why we don't maybe want to use spark when trying to serve delta data. We also discuss about using databases as query engines like duckdb.

Furthermore, the session discusses strategies for possible replication of the data or utilizing intermediary databases like Redis when low latency is essential. We consider factors such as business and application requirements and the challenges of possible data replication.

When it comes to serving Delta data through APIs, the discussion delves into frameworks like FastAPI and explores architectural choices such as serverless functions or containerization. It emphasizes the importance of simplicity and robustness in deployment. While not the fastest option, examples like the Databricks SQL API demonstrate that it can still be suitable for specific use cases where low latency isn't paramount.

Looking towards the future, the speech reflects on the ongoing trend of Delta Lake adoption and emerging projects built on this technology, while also pondering the accessibility of Delta Lake to all organizations. It raises concerns about the potential costs associated with cloud platforms and the learning curve of adopting new technologies like Delta Lake compared to more established solutions like MSSQL in the cloud.

In conclusion my goal with this session is to offer insights into the considerations and challenges involved in leveraging Delta Lake for serving data through APIs, encouraging thoughtful evaluation based on specific business needs, technical requirements, and cost considerations.