The course, in partnership with Dotex and taught by experts Wilk and Karen Fifer, focuses on techniques to enable language models (LMs) to produce structured outputs that can be easily read by other software components. This is particularly useful for extracting specific data fields for downstream applications. The course covers several approaches, starting with the use of proprietary APIs, such as those from OpenAI, which support structured outputs like JSON. These APIs allow for quick integration into AI applications, even though the internal workings are not always transparent. For models that do not natively support structured outputs or do so unreliably, the course introduces two main strategies. The first is the retry-brace method, which involves using libraries like Instructor to request a specific format from the model and retrying with additional hints if the initial attempt fails. The second strategy involves using the open-source library Outlines, which constrains the model during inference at the token level to adhere to a predefined schema or pattern. This ensures consistent adherence to format rules, eliminating the need for retries and enabling the generation of perfectly formatted outputs like JSON, CSV, HTML, or even structured game boards like tic-tac-toe. These techniques are crucial for ensuring reliable data output from LMs.