Breaktime Tech Talks podcast

Ep92: AI Pairing Pitfalls + RAG Benchmarking

0:00
15:31
Spola tillbaka 15 sekunder
Spola framåt 15 sekunder

In this episode, I give a whirlwind tour of upcoming events, share lessons learned from building an agentic Java application with AI coding tools, and dive into an academic paper on RAG benchmarking.

  • Upcoming events: All Things Open, JCON USA, Florida JUG tour, and NODES 2026 (Nov 11–12, free and virtual)
  • Road to NODES workshops: stopping agent hallucinations, graph reasoning, building AI that learns, context graphs & agent memory
  • Building an agentic Java app to streamline library book searches across multiple platforms
  • The over-optimization trap: choosing what to prioritize when AI expands what's possible in a short timeframe
  • AI-generated presentation outlines: good content, but missing the relatable storytelling that makes a session connect
  • The outsourcing framework: what to delegate to AI vs. what to keep in-house for creative work
  • The NICD RAG benchmarking paper: false positives/negatives, multi-hop reasoning, and safe refusal answers
  • Four pillars of RAG: question understanding, information retrieval, reasoning, and response synthesis
  • How the study modified traditional RAG metrics (answer relevancy, faithfulness, factual correctness) to reduce expensive LLM calls

As always, thanks for listening, and happy coding.

Fler avsnitt från "Breaktime Tech Talks"