How a .NET dev built an AI assistant

Short summary

A .NET developer shares lessons from building Cori, an AI assistant for 3D educational models, emphasizing vendor-neutral abstractions (Microsoft.Extensions.AI) over framework-specific APIs and solving the transport problem of delivering AI outputs to live clients. Key takeaway: abstract the LLM provider behind dependency injection and carefully architect how outputs reach your frontend in real-time.

• •Use vendor-neutral abstractions (Microsoft.Extensions.AI, IChatClient) to avoid lock-in to Semantic Kernel or other opinionated frameworks
• •Prioritize the transport/delivery problem (how AI outputs reach clients) over selecting the smartest model
• •Real-time streaming and tool invocation architecture matters more for UX than LLM selection alone

Generated with AI, which can make mistakes.