Stories, deep-dives and updates from the The Curator team.
AI Tool Calling: A Beginner’s Guide to Models That Can Act
Tool calling is the bridge between language models and operational software. Learn the vocabulary, execution loop, design trade-offs, and first safeguards required to turn generated instructions into dependable actions.
Read post
Inside Continuous Batching: How AI Servers Turn Uneven Requests into Efficient Inference
AI inference is not merely a model running on a GPU. Beneath the API, a scheduler continuously rearranges requests, memory, and compute. Understanding continuous batching reveals why throughput, latency, context length, and fairness remain inseparable product decisions.
Read post
Graph RAG vs. Vector RAG vs. SQL: Which Retrieval Architecture Should Ground Your AI Product?
Three retrieval architectures can connect an AI system to proprietary knowledge, but they preserve different kinds of truth. This comparison reveals when semantic similarity, explicit relationships, or structured queries should govern the answer.
Read post
How to Build a Capability Probe for an AI Model Before Designing the Product
A practical method for turning an untested AI model into an evidence-backed product decision through task sampling, blinded review, failure analysis, and boundary-setting.
Read postThree Myths About AI Evaluation That Let Weak Products Look Strong
AI evaluation is often mistaken for a leaderboard score, a single test set, or a final quality gate. The more useful reality is operational: evaluations must represent the work, expose costly failure modes, and continue after deployment.
Read postThe Compute Layer Is Becoming Programmable: Field Notes on AI Inference Routing
AI products are beginning to treat model selection as a runtime decision rather than a fixed architectural choice. The opportunity is not merely cheaper inference, but a control layer that can route each request according to difficulty, latency, privacy, and risk.
Read postEvent Sourcing: How to Build Software That Remembers Why Its State Changed
Most software stores only the latest state. Event sourcing preserves the sequence of decisions that produced it, creating systems that can explain, reconstruct, and reinterpret their own history.
Read postEmbeddings and Vector Search: A Beginner’s Guide to Software That Finds by Meaning
A practical orientation to embeddings, vector databases, hybrid retrieval, and the design choices that determine whether semantic search becomes useful infrastructure or an expensive illusion.
Read postInside Speculative Decoding: How AI Models Generate Faster Without Changing the Answer
Speculative decoding pairs a quick draft model with a stronger verifier to accelerate generation. The mechanism is elegant, but its value depends on acceptance rates, hardware behavior, sampling rules, and the shape of the workload.
Read postWhat The Curator blog is for
How it differs from the library
The library is the slow layer: ideas, methods and histories that should still read well next year. The blog is where the current argument lives — a launch worth interrogating, a strategy that keeps failing quietly, a piece of research that changes how a familiar problem looks. Posts argue; library entries explain.
Publication is paced rather than bulk-loaded, and every post is written specifically for this site, not shared across the sibling publications.
Editing and corrections
Posts are bylined and dated, and material post-publication changes are noted on the page rather than made quietly. Any reader can file a correction from an article page; the ones that check out are applied and credited.
External claims link to their source so readers can verify rather than trust. Our methodology page describes precisely where AI assists in drafting and where human editorial judgement is required before anything is published.