Example1: Rethinking API Discovery to Identify Duplicate API Services and Provide Visibility of Domain Mapping using BIAN and AI
The Issue: Over 30,000 APIs, both internal and external, have been established over the years by a large global bank in a variety of domains, including retail, wholesale, open, and corporate banking. There is a great chance that duplicate APIs will exist across domains, which will increase the total cost of ownership for managing the big portfolio of APIs and provide operational difficulties in handling API overlap and duplication.
Instead of locating pertinent APIs for reuse, API development teams create the same or comparable APIs due to a lack of visibility and discovery of the APIs. The Business and IT teams are unable to comprehend the capabilities that are already accessible and what new capabilities are required for the bank due to the inability to view the API portfolio from the perspective of the Banking Industry Model.
Generative AI-based method of solving problems: The solution learns the bank’s API portfolio and offers the capability to find APIs with auto-mapping to BIAN by utilizing the BERT Large Language Model, Sentence Transformer, Multiple Negatives Ranking Loss Function, and domain rules, all of which are tailored with knowledge of the BIAN Service Landscape. It links the API Endpoint Method to BIAN Service Operations, or level 4 of the BIAN Service Landscape Hierarchy.
The following are the main duties of solution:
- Consume swagger specifications and further API documentation to gain an understanding of the operations, endpoints, API, and related descriptions.
- Take in BIAN information and comprehend the BIAN Service Environment.
- Adjust with matching and erroneous mapping between BIAN Service Landscape and API Endpoint Method.
- With BIAN Hierarchical navigation and filters for BIAN levels, API Category, and matching score, provide the mapping and matching score visually.
Principal Advantages: With the aid of various filter and search options, developers were able to uncover reusable APIs based on BIAN business domains with ease. Teams were also able to pinpoint the most important API categories for correctly constructing operational resilience. The next iteration of the search function will be a conversational use case built on natural language.
A modernization approach that resolves and rationalizes duplicative capabilities was established with the assistance of the ability to discover duplicative APIs based on BIAN service domains.
The bank would have needed a year to find thousands of APIs before being able to accomplish the same outcome as this use case was done in just 7-8 weeks.
Example 2: MuleSoft APIs automated modernization to Java Spring Boot API
The issue: The sheer number of APIs, the dearth of documentation, and the complexity of the process were slowing down the current teams’ efforts to update MuleSoft APIs to Java Spring Boot.
Generative AI-based Solution Approach: IBM developed a Generative AI-based accelerator that greatly automated the Mule API to Java Spring boot modernization. IBM started by developing a thorough understanding of APIs, their parts, and their logic. Next, IBM finished writing the code and response structures. Next, building instructions were used to create Spring boot code that complies with MuleSoft’s API requirements, unit test cases, design document, and user interface using IBM’s version of Sidekick AI.
Using prompts, each Mule API component was entered into the tool individually. This produced an analogous Spring boot file, which was then wired together to fix any faults that arose. For the chosen channel, the accelerator produced a user interface (UI) that could be connected with the design documents, test data, unit test cases, and APIs. Sequence and class diagrams, requests, responses, endpoint information, error codes, and architectural concerns are all included in the design documentation that is produced.
Key Benefits: By combining a multi-model Generative AI technical strategy with technology and extensive domain understanding, Sidekick AI enhances Application Consultants’ daily work. The following are the main advantages:
- Produces the majority of the clean, optimized, and best-practice-compliant Spring Boot test cases and code; repeatability is crucial.
- Simplicity in integrating APIs with front-end channels.
- Developer’s code is easy to understand and has sufficient insight to be debugged.
In three sprints spanning six weeks, the Accelerator Proof of Concept was finished with four distinct scenarios involving code migration, unit test cases, design documentation, and UI development.
Due to a number of issues raised at the outset, including the time commitment required of SMEs, the impact of change on business, the need to modify operating models across security, change management, and numerous other organizations, many CIOs and CTOs have been reluctant to start modernization projects. Although generative AI is not a panacea for all issues, it does benefit the program by speeding up the modernization process, saving money on modernization, and above all de-risking the process by making sure no existing functionality is lost.
It is important to realize that integrating LLM Models and libraries with enterprise environment requirements such as extensive security and compliance evaluations and scanning takes time and work. In order to increase the quality of the data required for model tuning, some concentrated effort must be made. Although there aren’t any coherent Generative AI-driven modernization accelerators available currently, these integrated toolkits will eventually start to appear and will assist expedite at least some modernization trends.