Large Midwest Health Insurer
Streamlining Code Generation With Secure, Scalable GenAI
What if your GenAI solution could dramatically cut costs and empower developers to focus on higher-value innovation?
Challenge + Opportunity
Our client relied on a custom application to generate source code for data pipelines. To do this, the insurer sent source-to-target mapping (STM) files to an external large language model (LLM) API. Before being sent, sensitive patient data in the STM files had to be obfuscated. And when the API returned the generated code, developers had to manually reverse the obfuscation—an error-prone, time-consuming task that pulled them away from higher-value work.
In addition, the LLM used was not optimized for source code generation and came with high operational costs. The client aimed to reduce expenses, improve data security, and accelerate delivery timelines by modernizing its code-generation workflow.
Building an Automated Workflow With Databricks and LLMs
To secure and streamline the code-generation process, we designed and implemented a fully integrated workflow within Databricks. We hosted an open-source LLM that is specifically optimized for generating source code.
We tailored the model’s behavior through custom prompt engineering, enabling it to reliably interpret STM files and produce production-ready code with minimal post-processing. Exposing a secure endpoint within Databricks allows the entire process to run internally and eliminates the need to manually obfuscate and de-obfuscate data.
After resolving the obfuscation challenge, we tackled the broader inefficiency of manual code editing by automating the generation of ETL transformation and unit test scripts. Script generation from simplified input files now runs automatically to eliminate a manual bottleneck, accelerate development cycles in SQL and Scala environments, and reduce the risk of human error.
As a result of the shift to an open-source model, our client saw an 85% reduction in LLM inference costs. This work delivered an immediate bottom-line impact and created new capacity for strategic growth investments.
Results
Optimized Model Improves Security, Efficiency, and Costs
By keeping the entire workflow within Databricks, the solution supports enterprise-grade security and enables HIPAA-compliant data processing. This is an essential safeguard for patient privacy and reinforces regulatory confidence.
The model significantly improves performance by providing fast, low-latency responses. It also improves data accuracy and produces reliable, production-ready code that meets internal quality benchmarks. The new model increases productivity because it removes token limitations and reduces the amount of manual developer work.
Lastly, our client realized substantial cost savings. The shift to an open-source model reduced LLM inference costs by 85%. This delivered immediate bottom-line impact and created new capacity for strategic growth investments.
