Case Studies - Fransiscus Setiawan

Anonymized architecture notes from production software work across AI agents, Azure integration, .NET platforms, and EV charging systems.

AI Agent Workflow for Technical Operations

Problem: Engineers needed faster ways to inspect operational context spread across APIs, logs, and support data.
Architecture: A tool-calling agent with scoped MCP-style integrations, structured prompts, audit logging, and human review points.
Stack: .NET services, Azure-hosted APIs, secure secret handling, and model evaluation scripts.
Outcome: Reduced manual lookup time and created a repeatable pattern for safe agent access to internal tools.

Problem: Multiple systems needed reliable message exchange without point-to-point coupling.
Architecture: Azure Service Bus topics, dead-letter handling, retry policies, idempotent consumers, and operational dashboards.
Stack: C#, ASP.NET Core, Azure Service Bus, SQL, Azure DevOps.
Outcome: Improved reliability and gave teams a clearer support model for asynchronous workflows.

Problem: EV charging infrastructure required scalable charger communication, diagnostics, and partner integrations.
Architecture: OCPP message handling, charger state processing, integration APIs, and operational data pipelines.
Stack: .NET, Azure, SQL, event-driven services, REST APIs.
Outcome: Supported operational growth while keeping the platform observable and extensible.

Problem: Technical knowledge was scattered across documents, code, and historical tickets.
Architecture: Retrieval-augmented generation with curated sources, answer grounding, citation requirements, and feedback loops.
Stack: Azure AI Search, Azure OpenAI-compatible APIs, document processing, and .NET integration services.
Outcome: Created a foundation for answerable internal knowledge without bypassing source-of-truth systems.

Interested in the details behind any of these? Get in touch.