Code Harbor

Generative AI has tremendous potential for increasing coding efficiency. However, organizations often face significant challenges in its implementation - security, scalability, IP considerations and hallucinations EXL Code Harbor addresses these challenges using an extensive and reliable methodology.

The key steps of the process include:

Detailed Analysis: An under the hood analysis to understand the size, scale & complexity of the legacy codebase. It helps EXL to perform accurate effort estimation for the migration.

Explanation: Detailed documentation of each code into its natural language with descriptions of each function, step and key concepts of the code.

Data lineage: A pictorial, logical, and json view showing the flow of the logic from one table or variable over to another along with table and variable transformation logic tracked for both backward and forward lineage.

Data dictionary: Metadata creation mapping raw, intermediate and final derived tables along with variable format, description, transformation logic, source table and source variable for each variable within each table.

Chunking: Code pre-processing by splitting the source code into logical chunks, simplifying it for better quality conversion.

Conversion: Migrating source code to any language of choice, and then recompositing the code to get the full logic back.

Iterative Debugging: Agentic iterative debugging to solve for breaks and errors within the converted target code.

Optimization: Code optimization across code readability, syntax & function and logical improvements to enhance code performance, run time and memory usage.

Synthetic data generation: Test data creation either through reference data leveraging GANs or through a data dictionary in the absence of reference data to test and validate the converted target code output with source code output.

Testing: Creating test –and plan scenarios and generating the scripts to sufficiently test the code.

Output validation: Comparison of source code generated output with target code generated output across the suite of data validation and quality metrics. This human-in-the-loop step ensures complete input-output equivalency. Depending on the client's preference, it can be offered by EXL as a service or performed in-house by the client teams themselves.

Audit: Validate business logic and conditions present in codes to any structured or unstructured data.

Efficiency: Eliminate redundant code and enable performance optimization.

Documentation: Create comprehensive documentation to explain the coding process.

The tool requires S3, CloudWatch, Secrets Manager & ECR. The inferencing requires Sagemaker, ECS speaking to the Finetuned Model, along with RDS in the AWS tech stack. Multi-geography availability is obtained through Amazon ALB.

Tool Catalogue