I design and build AI agents that do real work inside real businesses. At Propel Finance I designed and rolled out an agentic software delivery pipeline, with AI agents and MCP services handling test-plan generation, build, and review and verification. Each stage is risk-rated with its own control gates, and development output is up 200%.
I'm now applying the same approach to legacy operations, replacing manual procedures on core finance systems with supervised, automated equivalents.
What I build
- Agents built on Anthropic Claude, Claude Code, OpenAI and AWS Bedrock
- Custom MCP (Model Context Protocol) servers that safely connect agents to your systems and data
- Agentic delivery pipelines for engineering teams: planning, build, test and review
- Back-office automation with humans in the loop where it matters
- Retrieval over your own documents and data
- Evaluation harnesses so you can measure quality before and after every change
Built to be trusted
Every agent I build has scoped permissions, audit logging, human approval gates for risky actions and clear fallbacks. Governance is designed in from day one, in line with ISO/IEC 42001.
What you get
- A working agent solving a real business problem, not a demo
- Secure integration with your existing platforms
- Monitoring, evaluation and cost controls
- Documentation and hand-over to your team