Enterprise AI Code Transformation Platform

What I personally owned

• Workflows for configuring, executing and monitoring AI-assisted code transformations.
• Source-code ingestion, project configuration and migration job creation interfaces.
• Transformation review, validation results and generated-code analysis interfaces.
• Dashboards presenting token usage, confidence scores and migration progress.
• Reusable accessible UI components, plus contributions to frontend architecture and coding standards.

Context

The product converts existing applications from one programming language to another without changing what the software does. A large language model performs the transformation, and a Python/FastAPI orchestration layer handles code parsing, LLM execution, validation and reporting.

The frontend is where that pipeline becomes usable: it is how a migration is set up, started, observed while it runs, and inspected once it produces output.

What made the frontend hard

A code migration is not a request/response interaction. It is a long-running job with partial results, retries, and outcomes that can be valid, invalid, or somewhere in between. The interface has to make an inherently uncertain process legible.

• Long-running jobs whose progress, errors and retries all have to be visible.
• Validation failures that are a normal outcome rather than an exception.
• Machine-generated output that engineers need to review before trusting it.
• Execution metadata arriving from several different backend systems.

What I owned

• Designed and implemented the configure / execute / monitor workflow for AI-assisted transformations.
• Built the source-code ingestion, project configuration and migration job creation screens.
• Built transformation review, validation results and generated-code analysis interfaces.
• Implemented dashboards for token usage, confidence scores and migration progress.
• Optimized rendering, state management, data fetching, frontend error handling and keyboard navigation.

Frontend architecture

The application is built with React, Next.js and TypeScript. I contributed to the frontend architecture, coding standards and component design, and built reusable accessible UI components that the migration workflows are assembled from.

• Typed component boundaries between configuration, execution and review surfaces.
• Shared accessible primitives rather than per-screen one-off components.
• State and data-fetching strategy tuned for screens that update while a job runs.

Asynchronous workflow design

Migration jobs run for a long time and fail in more than one way. The interface models progress, errors, retries and validation failures as first-class states rather than treating anything other than success as an error case.

• Explicit progress reporting while a transformation executes.
• Retry and error paths surfaced in the interface instead of hidden in logs.
• Validation failures presented as reviewable results.

Accessibility and performance

Accessibility work centred on reusable components and keyboard operation, so that the behaviour was consistent everywhere the components were used rather than fixed screen by screen.

• Reusable accessible UI components shared across the platform.
• Keyboard navigation treated as part of each workflow, not an afterthought.
• Rendering and data-fetching optimization for dashboards that update continuously.

Integration architecture

The React/Next.js interfaces integrate with Python/FastAPI orchestration services over REST for code parsing, LLM execution, validation and reporting. Execution metadata from MCP, Snowflake and Databricks pipelines is surfaced in the interface.

What was delivered

• End-to-end migration workflow: ingestion, configuration, job creation, execution, review.
• Dashboards for token usage, confidence scores and migration progress.
• A set of reusable accessible UI components used across the platform.