A specialist AI agent for enterprise software
I designed and deployed a specialist agent for an established enterprise application. I connected it to the existing engine so the same business rules apply whether a person starts a task directly or asks the agent to do it.
Tools and methods
Applied AI · Python · C# / .NET · Model Context Protocol · AWS
- Applied AI
- Python
- C# / .NET
- Model Context Protocol
- AWS
The interface between agent and engine
I wrote the agent in Python and designed and implemented the Model Context Protocol (MCP) server in C#. I defined which operations the agent could request, the inputs each needed and the structure of the result. That contract lets us extend what the agent can do without reimplementing business rules in Python.
Approval for changes to data
Looking up information and changing a customer’s data carry different risks. I designed an approval step for changes so the user could review exactly what the agent proposed before authorising it.
Deployment and ongoing development
I took the integration through deployment on AWS and verified the engine tools independently of the agent. That gave us a way to trace an unexpected result back to the application or the agent. I documented the contract for engineers adding new tools and shared findings with colleagues developing AI integrations elsewhere in the business.