I help businesses add useful AI capabilities to existing applications. From assistants that answer questions about company documents to tools that retrieve order information, I connect AI with application APIs, business rules, and user permissions.
The Business Problem
Many teams already have the data they need inside an application, ERP, CRM, or document library, but finding the right record or policy still takes time. A standalone chatbot cannot see that information safely. Useful AI integration works through the systems you already run, with the same rules and permissions your users have today.
Where AI Integration Helps
Internal support copilots
Help staff answer customer questions by looking up permitted records and relevant policies, then drafting a reply for a person to review.
Document-based question answering
Find relevant passages in company documents before generating an answer, so users can check where the answer came from.
Retrieving customer and order information
Let an assistant call existing application services to read the records a user is allowed to see, instead of querying the database freely.
AI-assisted steps in ERP and CRM workflows
Add suggestions, summaries, or lookups to an existing workflow through the system's available APIs and application services.
What an Integration Includes
Model and provider integration
Connecting your application to an LLM API, or to a local model through Ollama, behind an abstraction your code controls.
Retrieval over your documents
Chunking, embeddings, and semantic or hybrid search so answers are grounded in your own content.
Tools backed by normal .NET services
The model can request a defined tool; your application validates the request and runs its own business logic.
Server-side permission checks
User, role, and tenant rules are enforced by the application, never by the prompt or the browser.
Human confirmation where it matters
Draft replies and suggested actions are reviewed by a person before anything is sent or changed.
Read-Only Lookups vs. State-Changing Actions
Looking up an order status is very different from issuing a refund. I usually start with read-only tools that return permitted information. Any action that changes data goes through the application's existing service layer, with server-side validation and, where appropriate, explicit confirmation by a person. The model suggests; the application decides whether the action is allowed.
Demonstrated Project: SupportAgent.NET
SupportAgent.NET is my open-source ASP.NET Core and React project that shows this approach in practice: native tool calling over .NET business services, retrieval over company knowledge, and draft replies that remain under human review. It is a public demonstration project rather than a client deployment, and the source code is on GitHub.
Start With an AI Integration Assessment
A good first step is a focused assessment of one workflow. Together we review the workflow, identify the data sources and permissions involved, and outline an implementation plan. The outcome is a clear scope you can decide on before any larger build.
How I Work
- Discuss the workflow and the problem it should solve.
- Define data and permissions: which systems, documents, and actions are in scope.
- Build and test a focused implementation against real examples from your workflow.
- Review and hand over the result, with notes for your team.
Practical Limits
AI output can be wrong, so I design integrations that show sources, keep people in control of important decisions, and are tested on your own examples. AI is a tool to support your team, not a replacement for it, and the right approach depends on your data, privacy requirements, and running costs.
