Latest engineering dispatches — practical deep dives on architecture, performance, and tooling.
A floor-by-floor planning guide for self-hosting GLM-5.3-Flash: the ~306 GiB FP8 weight budget, why two H200s still cannot serve it, KV-cache sizing at the 1M-token window, and the realistic service-layer role for a home NAS.
A revenue-ops oriented guide to the Slack, Notion, and Salesforce MCP servers, including the documented Sales Call Follow-up template that automates post-call admin end-to-end.
A practical, threat-focused guide to MCP tool poisoning and rug pull attacks, covering how poisoned descriptions and post-approval tool swaps compromise servers, plus the input validation, schema enforcement, and output sanitization controls that stop them.
A cost-class taxonomy of ClickHouse ALTER operations, with explicit guidance on which ones are safe to run during business hours on MergeTree versus SharedMergeTree.
A walk-through of the entire Outlines pipeline that converts a Pydantic model into a token-level finite state machine, plus the formal-language-theory boundary that limits what regex-constrained generation can enforce.
A step-by-step dissection of one iteration of the harness loop, from prompt assembly through tool execution and observability, grounded in Claude Code and OpenDev internals.
A side-by-side comparison of MindsDB's federated SQL approach against traditional ETL pipelines for AI workloads, covering freshness, latency, operational cost, and the breakpoints where each architecture fails.
A side-by-side look at BERT's 512-token EHR splitting, LangExtract's multi-pass chunking, and Proxy-Pointer's structure-aware tree for structured extraction on documents that exceed any model's context window.
A practical walkthrough of how the Hugging Face Hub stores datasets — from Git repository layout and Xet chunking to the Google Cloud backbone underneath.
A technical teardown of the agent loop, context assembly, and tool wiring inside Claude Code, Cursor, and Codex — the machinery that makes modern coding agents work.
GLM-5.3 Flash is more than a cheap coding model; its native multimodal training makes it practical for video, 3D, CAD, slides, financial earnings, Godot games, vision-driven UI, and computer-use work. This guide walks through the recommended prompt patterns the Z.ai docs use for each non-coding…
A platform-engineering brief on why Redis 7.2 is at end-of-life, what the post-2024 licensing split means, and how to choose between Redis 8, Valkey, and Redis Software.
A troubleshooting-oriented look at five anti-patterns that quietly degrade LLM output — from stacking temperature and Top-P to misusing repetition penalties on reasoning models — with the reasoning and a safer alternative for each.
A practical walkthrough of Argilla's FeedbackDataset data model — Fields, Questions, Guidelines, and Records — aimed at engineers assembling preference data for DPO and RLHF.