leveraging_prompt_engineering_for_effective_openapi_descriptions_of_apis.pdf

Transcript

1. Leveraging Prompt Engineering for Effective OpenAPI Descriptions of APIs 1
2. codecentric AG Hoher Wall 15 44137 Dortmund Telefon: +49 (0) 151. 10 86 70 74 Daniel Kocot Head of API Experience & Operations AsyncAPI Ambassador daniel.kocot@codecentric.de www.codecentric.de Innovative - Trustful - Competent - Pragmatic 2
3. API (Description) Lifecycle 3
5. Focus on lifecycle steps
6. 6 Understanding the needs • API providers supported by API enablement teams need a deeper ramp-up in terms of design thinking methodologies. • So far, we rely on the use of personas to better understand the consumer base of the API. Based on the consumers, we can draw conclusions about their requirements. • In order to be able to look at these inferences in a meaningful way, the idea of the empathy map will be introduced. To further sharpen the results, the What- How-Why methodology will also be used. In the end, all this leads to a customer journey map. • With the help of AI, in particular ChatGPT, all results can be summarised in a mind map for a better overview.
7. Prompting… 7
8. Prompting… 8 • Create a developer persona with main technical skills are Java and OpenAPI.
9. Prompting… 9 • Create an empathy map for Alex Morgan.
10. Prompting… 10 • Please use What-How-Why methology with Alex Morgan.
11. Prompting… 11 • Gather all those informations above into a customer journey map.
12. 12 First draft of an API • When we develop a first draft of an API design, it always seems more important to write down the description directly using a specific definition language. • Instead, it is better to rely on a Domain Specific Language (DSL) that describes the models and a corresponding interface either based on a natural language or through paradigms and concepts of a programming language. • This leads to a corresponding trained large language model being able to fully support this process step.
13. Prompting… 13
14. Prompting… 14 • Using Copilot for Visual Studio • Describe an API via the chat or as comment in the code • Create a new product OpenAPI definition using problem+json for errors and HAL+json for the data. • Hopefully Copilot is able bring up the right data the right time • In Enterprise context it could be very helpful to create an LLM of existing API definitions
15. 15 Review the API • Here is an idea from @apihandyman to record the results of a validation as a basis for a possible review in an Excel table. • In order to draw conclusions about how APIs are designed within the organisation. • With regard to the evaluations, an AI could be used to provide a corresponding statement about the quality of the design of an API.
16. Prompting… 16
17. Prompting… 17 • Review the API definition. • Receving a protocol of • General Observations • Endpoint-specific Observations • Recommendations • Having a LLM on your own will be of good value for the organisation as a whole.
18. 18 Implement the API • With AI in combination with a portal or marketplace, among other things, we can make it clear to developers at an early stage which framework is suitable for their respective use case based on the description. • AI also reacts on the basis of data. • Once the framework or technology has been selected, the development process can be supported and driven forward with the help of Github Copilot.
19. Prompting… 19
20. Prompting… 20 • For this chatting with Copilot brings up the next steps to go depending on your framework decision.
21. 21 Test the API • For testing an API or the service, we can use AI in the sense of ChatGPT. • The test cases are created based on the respective definition and can be provided in different formats.
22. Prompting… 22
23. Prompting… 23 • Generate test cases.
24. Prompting… 24 • Generate Gherkin specifications from test cases above.
25. 25 codecentric AG | Hochstraße 11 | 42697 Solingen Creating the digital future together. Thank you!