Technical learning document
A short, opinionated path through the best spoken explanations of RAG — start with one clear clip, then hear how practitioners talk about grounding, chunking, and evaluation.
Curated follow-ups for RAG — open another explanation, then save what matters.
Save expert explanations into one investigation, compare voices, and export a shareable research brief on this device.
1. Start here
Chosen for clarity and how directly it answers the question — not for views or hype.
"How to build production ready RAG applications with Weaviate vector database"
RAG++ course: Hybrid search with Weaviate · Weights & Biases · 0:10
Opens a little earlier so you catch the setup
Share formats
A short, opinionated path through the best spoken explanations of RAG — start with one clear clip, then hear how practitioners talk about grounding, chunking, and evaluation.
retrieval → embeddings → chunking → vector DBs → reranking → evals → hallucinations
Grounding, context windows, and the retrieval-augmented loop.
Index, embed, search, rerank, generate.
Similarity search, indexes, and when dedicated vector DBs matter.
Hallucinations, wrong chunks, and ignored context.
Retrieval vs fine-tuning, chunking, evals.
Reranking, hybrid search, and production evals.
Save expert moments and export a team-ready brief.
Ground answers on wikis, runbooks, and tickets with clear source links.
Retrieve policy and product docs; watch for stale chunks after releases.
Compare expert explanations — save moments into a RAG investigation notebook.
Practitioners converge on these themes before debating tooling choices.
Open engineering debates — compare indexed explanations before you commit to an architecture.
Some experts prioritize retrieval for freshness and auditability; others invest in fine-tuning for stable domain tone and format.
Compare expert explanations →Fixed-size chunks versus semantic, structural, or agent-assisted splits with overlap tradeoffs.
Compare expert explanations →Whether bigger windows reduce the need for careful retrieval or only shift failure modes to attention and cost.
Compare expert explanations →When keyword/BM25 hybrid search is required versus dense embeddings alone for recall.
Compare expert explanations →Themes repeated across indexed engineering talks and practitioner writeups — not a survey, vote count, or attributed quote roundup.
Understand, then share
Save expert explanations into one investigation, compare voices, and export a shareable research brief on this device.
Use this when you need to explain RAG to someone else — save moments, compare voices, and export a brief they can read in Slack or Notion.
Supporting clips that deepen the guide's theme.
Matt Gotteiner: We've got Java and JavaScript SDK.
Why this is worth watching: Worth hearing after the opening clip.
Opens a little earlier so you catch the setup
Share formats
How practitioners frame real design choices — not a single “right” answer.
Most teams start with retrieval for freshness; fine-tune when vocabulary and style are stable.
Larger windows reduce chunking pain but increase cost and latency at query time.
Fine-tuning changes model weights; RAG retrieves external text at answer time.
Different experts and framings on the same topic — compare before you decide.
"a failure of OpenAI's training—where they have the intentions and they haven't met them yet— versus what is something th"
State of AI in 2026: LLMs, Coding, Scaling Laws, China, Agents, GPUs, AGI | Lex Fridman Podcast #490
Tutorial / walkthrough style
"models we have regarding the embedding and how to select the best"
LLM Fine-Tuning Course – From Supervised FT to RLHF, LoRA, and Multimodal
Tutorial / walkthrough style
"vector search is that we can take some input"
Agentic RAG: build a reasoning retrieval engine with Azure AI Search | BRK142
Tutorial / walkthrough style
"you actually optimize your rag"
Building Production-Ready RAG Applications: Jerry Liu
Technical / systems framing
Referenced by multiple experts — 4 distinct channels in this comparison.
A polished answer on wrong chunks fails silently — measure retrieval recall first.
Offline benchmarks miss drift; log failure modes where answers ignore retrieved text.
Retrieval-augmented generation (RAG) grounds a language model on retrieved documents at query time. The clearest expert
RAG hallucinations often come from wrong or missing chunks — not from the model “making things up” in isolation. Experts
Teams evaluate RAG in two layers: retrieval (did we fetch the right chunks?) and generation (did the answer stay faithfu
A vector database stores embeddings for similarity search; RAG is the full pipeline that retrieves passages and conditio
Chunking splits documents before embedding and retrieval. Experts warn that fixed-size splits, missing metadata boundari
RAG observability traces retrieval, context assembly, and generation so teams can see which chunks were shown, whether r
Side-by-side architecture, retrieval, and framework tradeoffs with expert context.
RAG updates what the model can read at query time when facts change; fine-tuning updates how the mod
Semantic search returns ranked passages by embedding similarity. RAG adds chunking strategy, context
Chunking determines which text exists in the index at all. Reranking only reorders candidates alread
Pinecone optimizes for managed approximate nearest-neighbor search with minimal ops. Weaviate offers
Agents plan and execute multi-step workflows with tools. RAG measures whether the right text was ret
RAG retrieves relevant text at query time, then generates an answer grounded on that context. Practi
Chunking defines the searchable units in your index. Size, overlap, and structure-aware splits deter
Practitioners prioritize retrieval recall on required facts per question set before generation metri
From retrieval basics through indexes, reranking, and evaluation.
Contrasting explanations from long-form talks — use both sides to stress-test your design, not to pick a winner.
Teams disagree on when to retrieve context versus adapt model weights.
Keep weights frozen; ground answers with external chunks.
Adapt the model when domain vocabulary is fixed and labeled data exists.
Chunk size and overlap change recall and answer quality in different ways.
Higher recall for precise facts; more index noise.
Better narrative context; risk missing needle facts.
Some engineers ship hybrid search; others rely on dedicated vector stores.
Scale ANN indexes and metadata filters separately.
Combine BM25 with embeddings in one stack.
Sequenced RAG engineering path — each link builds on the last concept, not random suggestions.
Open expert guide →