← All workFinancial Services · Modernization

AI that reverse-engineers legacy code into specifications

A GenAI pipeline using GitHub Copilot and Claude that turns undocumented legacy codebases into functional and technical specifications — then a forward-engineering platform that builds from them.

Role
Solution Architect & AI Delivery Lead
Period
2023 – 2026
Stack
ClaudeGitHub CopilotCopilot agentsSource controlTest scaffolding
Legacy applications documented
5+
Analyst effort on discovery
−30%
Build cycle time (forward engineering)
−50%

Context

Modernization programs stall at discovery. The people who wrote the legacy system have moved on, the documentation is out of date, and analysts spend weeks reading code to work out what it actually does.

Reverse engineering

A GenAI pipeline built on GitHub Copilot and Claude reads legacy codebases and produces functional and technical specifications: what each module does, the business rules it encodes, and how the pieces call one another. Analysts review and correct the output rather than writing it from scratch.

Applied across 5+ legacy applications, it cut analyst effort by 30%.

Forward engineering

The specifications then fed a second platform — GitHub Copilot and Claude agents that automate code generation, test scaffolding and bug resolution from functional specs — cutting build cycle time by 50%.

What made it work

  • Humans review every specification. The model drafts; the analyst signs off.
  • Specs live in source control, so they evolve with the code instead of going stale.
  • Specs before code. Generating from a reviewed specification is far more reliable than generating from a vague ticket.

Client identity withheld. Details are limited to what can be shared publicly.