Navigating Challenges and Enhancing Performance of LLM based Applications
summary
Covers practical RAG specifics: ingestion→retrieval→synthesis→generation, use of vector DBs (Qdrant), chunking tradeoffs and the lost-in-the-middle problem, and concrete levers that moved metrics—switching embedding models, changing retrieval window size (2–5 chunks), HNSW vs lexical indexes, reranking and multimodal retrieval. Cites adoption stats (~60–65% RAG use cases) and an embedding leaderboard with 159 datasets, 113 languages, 310 models. Publishes notebooks and a git repo showing that embedding choice plus reranking materially improves relevance and reduces hallucination when paired with OpenAI / Mistral / Llama 3 models.
AI-written summary · about
chapters (12)
- 0:00Introduction and speaker intro
- 1:16Why LLMs are trending now
- 2:44What is RAG and its challenges
- 7:14Key challenges in LLM applications
- 11:50Improving RAG performance
- 13:57Evaluation framework for LLMs
- 15:20Model evaluation strategies
- 17:28Data chunking and processing
- 20:20Retrieval evaluation methods
- 22:49LLM response evaluation
- 25:16Tools and frameworks for evaluation
- 27:19Summary and resources
related talks
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
19mRAG in 2025: State of the Art and the Road Forward
30mRAG: Tricks from the trenches.
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
53mFrom RAG to Riches: Secrets of Data-Augmented LLMs
57mLLMs gone wild
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
56mRAG: from dumb implementation to serious results