Natural-language interface for asset, order, and employee operations.
MCP server + AI agent that turned multi-step internal tools into a single conversation.
An AI assistant layered on top of an internal asset management system. Users manage assets, asset orders, and employees through natural language instead of navigating multiple screens.
Internal teams were losing time navigating multi-step CRUD flows. Naive LLM integrations were slow, token-heavy, and unreliable on real production data.
Built a Model Context Protocol (MCP) server exposing typed tools for assets, orders, and employees. Engineered prompts and context windows carefully — only the data the model needs, in the shape it needs it.
Reached ~95% task accuracy on real internal workflows with ~60% fewer tokens per request compared to the initial implementation, through prompt and context optimization.
- Designed and built the MCP server and tool schema
- Implemented the AI agent loop with structured tool calling
- Engineered prompts and context windows for accuracy and cost
- Reduced token usage by ~60% through context optimization
- Validated outputs against production data for safety
Shows applied AI engineering: real outcomes, measurable accuracy, and disciplined cost control — not buzzwords.