AI-assisted modernisation
Use AI to read, document, test and convert legacy code faster, with engineers reviewing every change and your code kept under your control.
Faster understanding, safer change
The slowest part of modernising an old system is understanding what its code actually does. Large language models are now good at exactly that: summarising modules, tracing business rules, explaining unfamiliar languages and drafting tests and translations.
We use AI as an accelerator inside a disciplined engineering process. AI drafts; engineers decide. Every output is reviewed, tested and traceable, and the models run only where your security policy allows.
When it fits
01
Large codebases, little documentation
Hundreds of thousands of lines, written over decades by people who have since left.
02
Rare languages and skills
COBOL, PL/SQL, PowerBuilder or Visual Basic that few engineers can still read quickly.
03
Almost no automated tests
So every change is a gamble and every release needs weeks of manual checking.
04
Large conversion programmes
Where converting by hand alone would take years and cost more than the business case allows.
What we do
Code comprehension
Summaries of modules, call graphs and data flows, so engineers can find their way around unfamiliar code in days.
Business rule extraction
Rules found in code, triggers and stored procedures, written up in plain language for business owners to confirm.
Documentation
Architecture notes, API descriptions and runbooks drafted from the code and kept alongside it.
Test generation
Unit and characterisation tests drafted from existing behaviour, so changes can be made safely.
Code translation and refactoring
Draft translations from legacy languages and frameworks to modern ones, refined and tested by engineers.
Private, governed AI
Models deployed where your policy allows, including on infrastructure you control, with access control and logging.
Human in the loop
AI drafts. Engineers decide.
AI output is a draft, never a commit. Each suggestion passes engineer review, automated tests and the same quality gates as code written by hand. Drafts that don't pass go back to be redone.
- AI drafts
- Documentation, tests and code, generated from the legacy source.
- Engineer review
- An engineer checks every draft, and rejected drafts are redone.
- Tests, then merge
- Only reviewed, tested work reaches the repository.
How we use AI on a modernisation
01
Agree the guardrails
Decide which code, models and environments are allowed, and how outputs are reviewed and recorded.
02
Index the codebase
Load code, schemas and documents into a private workspace the AI tools can search and reason over.
03
Accelerate the work
Use AI to explain, document, test and translate, with engineers checking every result.
04
Measure and improve
Track time saved, defects found and review effort, and tune prompts and tools as the work goes on.
What you get
- An AI usage policy and guardrails for the programme
- A searchable knowledge base of the codebase
- Reviewed documentation for modules and interfaces
- A business rule catalogue confirmed by owners
- Characterisation and unit test suites
- Translated or refactored code, reviewed and tested
Platforms and tools
- Enterprise coding assistants
- Open-weight models hosted privately
- vLLM
- NVIDIA NIM
- Retrieval over code and documents
- Code graphs and static analysis
- SonarQube
- Automated test frameworks
Questions about AI-assisted modernisation
Will AI replace the engineers on the project?
No. It removes the slowest, most repetitive work, so engineers spend their time on decisions, review and testing.
Is our code sent to a public AI service?
Only if you agree. We can use enterprise services that don't retain or train on your data, or models hosted on infrastructure you control.
How accurate is AI-generated code?
Accurate enough to save a great deal of time, not accurate enough to trust unchecked. That's why every output goes through review and automated tests.
Can AI convert COBOL or PL/SQL automatically?
It produces useful first drafts, especially with good tests around them. Complex, critical logic still needs careful engineering, and we plan for it.
See what AI can do for your codebase
Tell us about the system, its languages and its size. We can run a short pilot on one module to measure what AI assistance saves in practice.