Automating RedHawk-SC Map Reduce Query Code Generation and Interactive Q&A using Large Language Models

"Large Language Models (LLMs) such as ChatGPT, BARD, Llama etc., have recently exploded in the world of generative AI with a very wide variety of novel applications. These models have demonstrated amazing capabilities in a range of tasks such as text generation, knowledge extraction, summarization, code generation, problem solving and providing suggestions. These LLMs can also be adapted or tailored to specific applications through fine tuning making them quite versatile.

One of the key challenges in RedHawk-SC usage by the designers is writing map reduce code in Python for performing various queries on a range of analyses. These queries can sometimes become quite complex and designers not having a programming background may find it difficult and inefficient to write such code and need help from Ansys AEs and product specialists. This work attempts to automate the Python map-reduce code generation by adapting and fine tuning an LLM using the large set of examples that our team has created over the years. This will enable users of RedHawk-SC to interact with the adapted LLM in natural language to generate the required Python map reduce code.

Another use case that this work will address is providing interactive Q&A on the RedHawk-SC usage, analyses and debugging. The LLM will be fine-tuned on a knowledge base from a variety of resources such as user manuals, white papers, application notes, discussion channels etc. This will enable the users of the tool to interactively get answers to their questions, get debugging hints, suggestions on tool usage and analyses.

We envisage that such LLM applications will grow at Ansys and provide significant productivity improvements for both Ansys and its customers.

"