Extending an open source LLM to support efficient code generation in intersystems technology

Description of the idea

1. Currently, source code generating LLMs generate ObjectScript/BPL/DTL/PEX/SQL/MDX/CSP/XML for Productions/BI and Embedded Python code in a limited and/or inaccurate manner.

2. InterSystems could specialize an already trained open source code generation model (CodeLlama, StarCoder, etc.) and include support for InterSystems IRIS languages ​​and technologies.

3. If possible, it could publish the result as an open Docker container for the community, making life easier especially for new customers who are not yet familiar with IRIS and thus increasing customer retention.

Who is the target audience?

1. InterSystems IRIS Developers

2. Potential clients testing and trying IRIS

What problem does it solve?

1. Deliver project on time, with quality and on budget

2. Improve learning curve

3. Be more low/no code

4. Reduce development complexity

How does this impact the efficiency, stability, reliability, etc, of the product?

1. Improve product UX

2. Ease to use

3. Product will be more low/no code

4. Source code more secure, stable, reliable and efficient

Provide a specific use case or scenario that illustrates how this idea could be used in practice.

1. I am a potencial client testing main players (IBM, Oracle, Microsoft, InterSystems) solutions. I am looking for solutions to create software with productivity, quality and reducing costs. With this solution, the developer can use a prompt to generate an end-to-end interoperability solution beetween Salesforce and SAP with Python PEX and IRIS SAP connector. In 2 to 5 minutes, the developer has 80% to 90% of the source code ready and documented.