Notes from the work
Build the thing.
Explain the hard part.
Practical tutorials about AI agents and modern software systems—tested against real constraints, written so you can make a better next decision.
- Evidence
- Working code, observable results, and primary sources.
- Judgment
- Tradeoffs, failure modes, and when not to use the technique.
- Freshness
- Every tutorial shows when it was reviewed—and a tested date only when its example was run.
Start here
Three useful ways in.
A short editorial shelf for readers arriving with a problem to solve.
AI agents · Multi-agent systems · Architecture
Agent Swarms: When Multiple AI Agents Help—and When They Make Things Worse
ReadAI agent memory · PostgreSQL · pgvector
AI Agent Memory with Postgres and pgvector: Schema, Retrieval, and Forgetting
ReadModel Context Protocol · TypeScript · AI agents
Build a Safe MCP Server in TypeScript: Tools, Validation, and Transports
Read
The archive
Useful past the quickstart.
Showing 1–12 of 107 archive entries.
- Read the tutorial
Accessibility · AI agents · User experience
Accessible Agent Interfaces for Keyboard, Screen Reader, and Voice Users
Design streaming conversations, tool progress, approvals, errors, and voice controls that preserve focus, user control, and understandable status.
- Read the tutorial
OpenTelemetry · AI observability · Semantic conventions
Adopting OpenTelemetry GenAI Semantic Conventions Without Breaking Observability
A migration playbook for adopting the evolving OpenTelemetry GenAI span, metric, and event schema safely.
- Read the tutorial
AI agent evaluation · Continuous integration · AI testing
AI Agent Evals in CI: A Release Gate for Prompts, Models, and Tools
Build a provider-neutral agent evaluation suite with deterministic checks, repeated trials, safety cases, uncertainty, and a CI policy that catches regressions.
- Read the tutorial
AI security · AI agent memory · Prompt injection
AI Agent Memory Poisoning: Threat Model, Detection, and Recovery
Protect persistent agent memory with source trust, write gates, quarantine, contradiction checks, provenance, and complete recovery procedures.
- Read the tutorial
vLLM · Prefix caching · AI agents
Automatic Prefix Caching for Agent Workloads
Make repeated agent prompts cacheable in vLLM without crossing tenant boundaries, and measure whether cache hits actually improve latency.
- Read the tutorial
AI agents · Reliability · Distributed systems
Backpressure for AI Agent Systems
Design admission control, bounded queues, concurrency limits, deadlines, and graceful degradation for agent workloads.
- Read the tutorial
PostgreSQL · RAG · Benchmarking
Benchmark PostgreSQL 18 Async I/O for RAG Workloads
A reproducible protocol for testing PostgreSQL 18 AIO across vector search, filtered retrieval, ingestion, vacuum, and cache states.
- Read the tutorial
Quantization · LLM inference · Benchmarking
Benchmark Quantized Agent Models for Quality, Memory, and Speed
A reproducible protocol for comparing weight, activation, and KV-cache quantization without inventing savings or hiding quality regressions.
- Read the tutorial
Speculative decoding · vLLM · Benchmarking
Benchmark Speculative Decoding Reproducibly
Measure speculative decoding against a target-only baseline across load, prompt types, acceptance rates, memory, latency, throughput, and output equivalence.
- Read the tutorial
Structured outputs · Open models · Evaluation
Benchmark Structured-Output Reliability in Open Models
A reproducible protocol for measuring JSON validity, schema compliance, semantic accuracy, latency, and failure recovery across open models.
- Read the tutorial
AI agents · Cost control · Reliability
Build a Budget Controller for Agent Workflows
Enforce token, tool-call, latency, concurrency, and monetary ceilings across nested agent work without relying on prompt instructions.
- Read the tutorial
Multimodal AI · Document AI · Retrieval-augmented generation
Build a Citation-Aware Document Agent for PDFs, Scans, Tables, and Images
Preserve page regions, reading order, tables, OCR alternatives, and source versions so every document answer can resolve to visible evidence.