Reusable ML Templates and Training
Case study
Outcome
- Built reusable ML frameworks and coding templates that cut project onboarding time by 90% and standardised working practices across consulting teams.
- Worked as the AI subject matter expert across credit risk and insurance, advising on ML strategy and AutoML and building proofs of concept.
- Ran monthly sessions and created learning material on Python, ML and software engineering for more than 100 consultants.
- Developed ML frameworks covering data pipelines, model templates, evaluation utilities and deployment scripts.
Context
When I joined 4most, the business was still heavily SAS-based, while the wider banking and financial services industry was beginning a significant shift towards Python. As one of the company’s early Python power users, I helped accelerate that transition internally while supporting client-facing data science delivery.
Challenge
My internal focus was reducing the time it took consultants to become productive on Python-based projects. I supported this through Git and version control training, reusable Python notebook templates, and practical upskilling initiatives for consultants between client engagements.
Approach
The templates covered common data science tasks that consultants encountered on client projects, with explanatory markdown and reusable code. Other consultants added templates while practising agile workflows; I reviewed their pull requests and maintained the library.
Alongside the internal work, I joined financial services and insurance clients to review early Python implementations, improve modelling workflows and add to existing systems. I also advised teams on development practices and ran workshops where they were useful.
Results
Project onboarding
- Implemented husky across all internal repositories to enforce coding standards
- Built a template library covering common ML project workflows
- Worked with IT to move Python access to uv to meet project requirements and manage version vulnerabilities
- Built an upskilling programme for consultants between client projects
- Reduced project onboarding time by 90%
AI subject matter expert
- Advised credit risk and insurance teams on ML strategy and AutoML adoption
- Developed AI roadmaps with clients
- Built proofs of concept to test proposed AI use cases
- Carried out technical due diligence and architecture reviews
Reusable frameworks
- Data preprocessing and feature engineering pipelines
- Model training and evaluation templates
- Graphics modules for presentations