The retrieval layer is where a company's AI gets its memory.
Field notes on building the retrieval layer under a company's AI: how a RAG system goes together, what chunking and embedding choices do to answer quality, how to evaluate a system honestly, and where permission-aware retrieval becomes the security boundary. Written for the engineers who will operate it, not for the person signing the contract.
Retrieval is an engineering discipline, not a library call.
This area is about the layer between your documents and the model: how a RAG system goes together, what chunking and embedding choices do to answer quality, how to evaluate a system honestly, and where permission-aware retrieval becomes the security boundary. It is written for the engineers and architects who will be paged when the assistant answers from the wrong document.
The position we take here is that retrieval quality dominates model choice. A well-chunked, well-evaluated pipeline on a mid-size open-weight model beats a frontier model bolted onto bad retrieval, and the difference shows up in the questions users stop asking. Most of the interesting failure modes in enterprise RAG are retrieval failures wearing a model failure's clothes.
The enterprise constraint changes the architecture rather than just the deployment. Permission-aware retrieval — the index knows who may read what, and the search respects it — is what turns a helpful assistant into one a compliance officer can sign off. The posts here spend their time on that layer.
Posts on retrieval and RAG
agentic ai
Agent Engineering: 7 Production Skills Beyond Prompting
Prompt engineering is the recipe; agent engineering is the cook. The 7 skills — system design, tool contracts, retrieval, reliability, security, metrics.
AI Workflows vs AI Agents: Who Decides — and When It Matters
AI workflows follow a fixed path a human defines; AI agents let the model decide. RAG is a workflow, ReAct is the agent pattern. When each is the right call.
Scaling AI Agents: Multi-Agent Design Is the Answer
Scaling agents is not about more infrastructure. Errors propagate, costs grow nonlinearly, and one agent owning everything becomes the bottleneck.
rag
How to Build a RAG System: The Complete Guide
RAG system build guide: chunking strategies, embedding models, vector search, retrieval quality, evaluation, and the rollout that survives a security review.
RAG vs Fine-Tuning vs Prompt Engineering: The Decision Matrix
RAG, fine-tuning and prompt engineering solve different problems. The three-way matrix shows which fits current data, stable patterns or plain instructions.
Common questions
What is the company brain and why does it need a retrieval layer?
The company brain is the internal knowledge base a company's AI assistants answer from: contracts, engineering documents, quality records, production data. Retrieval is the layer that finds the relevant text before the model answers, and its quality decides whether the assistant is trusted or switched off.
Is this area about RAG?
Mostly. Retrieval-augmented generation is the standard architecture for answering from your own documents without retraining a model, and it is where the interesting engineering decisions are: chunking strategy, embedding models, vector store choice, retrieval quality and evaluation. The posts here take positions on those decisions.
Not a sales call. An architecture call.
Thirty minutes with the architect who would actually run the engagement.