RAG Against the Machine
summary
Practical RAG blueprint that builds a self‑hosted pipeline using a sentence embedding model, Chroma vector DB, LangChain orchestration and an open‑source LLM (nomic / GPT4All), with a Chroma vector DB backend and MITRE-embed demo that maps Splunk/Sigma detections to MITRE techniques. Benchmarks aim for 10s response target and under $500/month on a G4dn GPU, analysts (~5 users) generate ~5,000 tokens/day each (100k tokens/day aggregate). Live demos include Slack integration and policy gap analysis against UN regulatory texts, and prototypes succeed in clustering, retrieval and QA.
AI-written summary · about
chapters (21)
- 0:00RAG in cyber security
- 1:10Agenda
- 2:56Speaker's expertise and industry experience
- 4:23AI in cyber security
- 9:20AI guide not silver bullet
- 10:44Retrieval augmented generation
- 13:37Build RAG system from scratch
- 17:08Visualize data with Spotlight
- 18:54Data workflow architecture
- 20:23Security implications of RAG
- 21:25User interaction cost benchmarking
- 23:13Phoenix project AI integration
- 23:55Security controls for RAG
- 26:47MITRE embed demo
- 33:10Threat intelligence use cases
- 37:48Cost transparency AI efficiency
- 43:53Laws regulations with RAG
- 49:09Regulatory compliance mapping
- 51:38RAG with Slack integration
- 52:43Next steps research
- 56:20Q&A
related talks
25mSet Up an Open Source RAG in 30 Minutes
41mFrom RAGs to riches: Build an AI document interrogation app in 30 mins
53mBuild RAG from Scratch
42mTake Your RAG To The Max!
41mMerging Human Cyber Security Expertise with Generative AI
01h 13mBuilding Production-Ready AI Systems in 90 Minutes
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
56mRAG: from dumb implementation to serious results