Web Development Then, LLMs Now: Forecasting LLM Best Practices by Looking Back – and Ahead

I gave a talk at the 7CTOs Colloquium in Austin on best practices for building Retrieval-Augmented Generation (RAG) pipelines—and how we’re starting to see a clear architecture emerge for serious LLM-based applications.

In this talk, I cover:

• Chunking strategies (simple, structural, semantic)
• Search and retrieval (BM25, vector search, RRF)
• Query rewriting, decomposition, and HyDE
• Reranking, repacking, and summarization
• Tool calling and query classification

I also share prompts, techniques, and design tradeoffs we’ve discovered while building real-world LLM applications at Scientist.com.

Thanks to 7CTOs for putting on the event and thank you to Howdy.com for hosting!