.NET Rocks! Agentic RAG with Ed Charbeneau
14 snips
Apr 8, 2026 Ed Charbeneau, principal developer advocate at Progress who builds .NET SDKs and AI demos. He discusses making agents smarter with Progress Agentic RAG and NucliaDB as a vector data store. Topics include RAG basics, ingestion types (PDFs, audio, web), extraction of structured data, hybrid search techniques, monitoring quality, scalability, and .NET developer experience.
AI Snips
Chapters
Transcript
Episode notes
How RAG Augments LLMs With Document Context
- Retrieval Augmented Generation (RAG) augments an LLM by injecting relevant retrieved document context before prompting the model.
- Ed Charbeneau explains chunking documents, creating embeddings, and using semantic (vector) search to return meaningful context for the model.
Don't Roll Your Own RAG Plumbing
- Avoid rolling your own RAG plumbing unless you have dedicated ML/data-science resources because each piece (models, embeddings, DB, ingestion) has complexity and scale costs.
- Ed recommends using a prescribed RAG-as-a-service to speed time-to-value and reduce infrastructure work.
Use Hybrid Deployment For Sensitive Data
- Choose hybrid deployment for sensitive data: host the vector DB on-prem and run models in cloud or use fully on-prem models like Llama when required.
- Ed emphasizes plug-and-play model endpoints and custom keys for flexible deployment.