Eradani Taps GenAI for IBM i Integration Boost

October 9, 2024 Alex Woodie

Eradani introduced a new product last month that uses the power of generative AI to simplify the development of code for integration projects involving IBM i. The company says the new product, dubbed Eradani Assist, will allow customers to use natural language to tell the product to build specific application and data integrations, and the product will leverage one or more large language models (LLMs) to generate the integration code.

Eradani calls its flagship product, Eradani Connect, an “integration hub” that manages communication into and out of the IBM i platform. The software supports application and data integration projects that utilize a range of protocols and techniques, including APIs using both REST and SOAP; data exchange using EDI standards; batch data movement using FTP/sFTP support; and real-time event-driven Pub/Sub frameworks, like Apache Kafka.

The goal with Eradani Connect is to make the IBM i look like just another standards-based application and data repository in the corporate network, says Eradani CEO Dan Magid. Instead of forcing developers to learn the intricacies of accessing data and applications on IBM i, Eradani Connect allows them to treat IBM i like any modern endpoint that communicates using standard protocols, while it handles the IBM i-specific connections under the covers.

“So if I am a. NET developer, if I’m a Java developer, or a JavaScript developer, and I need to get something from the IBM i, I can get to the IBM i in the way I normally would call to any other platform, and we will take care of all the stuff that needs to get done in order to call an RPG program, or to take JSON or XML data and turn it into RPG data structures and parameter data,” Magid says.

The company is applying the same principles of striving for simplicity and shielding users from complexity with Eradani Assist, which the company unveiled three weeks ago during the COMMON NAViGATE conference in Bonita Springs, Florida.

According to Eradani, the new product will automatically generate industry much of the standard code for application and data integration involving IBM i. The software leverage the power of publicly accessible large language models (LLMs) to automatically generate the boilerplate JavaScript code that’s necessary for integration workflows, which will save time and boost productivity for developers. Because LLMs do not yet understand RPG, Eradani’s uses its own proprietary technology for generating RPG.

Eradani Assist allows users to generate application and data integration workflows using natural language.

Eradani Assist (currently in beta) is all about saving time for the developer by handling basic but tedious tasks that otherwise would have to be done manually, Magid says.

“They just go in and say in English what is it that they want it to do,” he says. “Under the covers, we’re going to do all the hard stuff of setting up the authentication, of setting up the routing, of setting up the data transformations, if it has to get transformed from IBM i data to JSON or XML data, or from comma delimited files to IBM i data structures. We’re going to do all that stuff under the covers so that they don’t have to deal with it.”

During a demo, the CEO used Eradani Assist to develop an API call for a latitude and longitude lookup using Eradani Connect running as a plug-in for VS Code. The product started out by generating a stub for the API that specified how the integration would be configured, including authentication settings, encryption, etc. Magid provided the product with an API key for the lat/long lookup, and Eradani Assist generated the outbound API call around it.

When it was finished generating the code for the lat/long lookup, Magid then told Eradani Assist to perform a temperature lookup based on the lat/long results. The product successfully completed that task, even though it wasn’t told specifically what weather API to use. The LLM that Eradani Assist is using – Claude from Anthropic, in this case – figured out on its own which weather API to use, Magid says. “The LLM just chose one,” he says.

That demonstrates one of the interesting aspects of Eradani Assist and GenAI in general: users can be as specific as they want with Eradani Assist, or not specific at all. If users have documented their APIs with a 20-page Swagger document, Eradani Assist can use that information in the prompt that it sends to the LLM.

That capability to be more specific and exact will come in handy with more complex integration tasks. For instance, if a customer has a large document describing their specific Electronic Document Interchange (EDI) mappings, Eradani Assist can use that information to generate new EDI connections, says Aaron Magid, the vice president of open source technologies at Eradani and Dan Magid’s son.

“Where that really becomes important is using that for more complex things like EDI integrations where we’re saying, okay, I’ve got this EDI document, I’ve got the EDI standard, I’ve got this huge PDF document, I’ve got some samples here and there, it all comes together to be a to be a huge amount of information,” the younger Magid says. “You put it all in there and say, map everything and build an integration for me that will process this data. And it goes out and it and it gets it.”

LLMs like Claude perform quite well with well-documented languages, such as JavaScript. Aaron Magid has been writing JavaScript for 15 years and he is a big believer in using GenAI to generate boilerplate code.

“I could write all this. It’s native to me, but it’s so much faster and so much easier” to leave it to GenAI, he says. “The cognitive drain of trying to figure out how to call a weather API and get all that – it adds up when you’re coding.”

That’s not to say that Eradani Assist will never make a mistake. Just like a junior developer will perform well at most tasks but occasionally make an error, Eradani Assist should be monitored closely by humans to make sure that it’s not hallucinating something and going off the rails. To that end, Eradani developed the software to create comments as it’s generating the code, and also to monitor the code that the software generates.

Achieving that visibility into Eradani Assist code it easier since it generates industry-standard JavaScript code. However, the company’s initial idea for using GenAI was to create a low-code environment, where the user drags and drops things around on the screen, and then the product generates the integration code under the covers. The potential for vendor lock-in created by low-code environments led Dan Magid and company to take a different approach.

“We looked at the large language models and said, well, instead of doing the drag-and-drop, low-code thing, let’s just let the customer describe what they want, we’ll generate standard RPG code, standard JavaScript code, and then if they want to, they can they can modify the code,” the elder Magid says.

While calls to LLMs are used to generate the integration code, no LLMs are involved when the integrate code is actually executed, the company says.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search