AI Wisdom
📝 CHANGELOG

Changelog

Weekly updates on what's new across the AI Wisdom platform — new sections, content drops, and engineering improvements.

Reference Layer Launch — Glossary, Patterns, Decisions & Changelog

This week we shipped the AI Wisdom Reference Layer — four new sections designed to be your quick-lookup companion while reading articles or building AI systems.

What's new:

  • ▸Glossary — 30 core AI engineering terms with concise definitions and deep-dive explanations. Every term links to related articles and other glossary entries.
  • ▸Patterns — 5 production-tested AI engineering patterns (Semantic Cache, Intent Router, Token Budget Gate, Tiered Retrieval, Tenant Isolator) with problem/solution format, trade-offs, and when-to-use guidance.
  • ▸Decisions (ADRs) — 5 Architecture Decision Records documenting real choices we made building AI Wisdom: RAG vs Fine-Tuning, Vector DB selection, model selection, orchestration framework, and embedding model choice.
  • ▸Changelog — You're reading it! Weekly updates on what's new across the platform.
  • ▸Reference dropdown added to the main navigation for easy access to all reference sections.

All reference content is fully integrated into the sitemap for SEO discoverability.

New Domains: TypeScript Mastery, Algorithms & Infrastructure as Code

Three new engineering domains shipped this week, bringing deep-dive content for engineers who want to level up core software skills alongside AI development.

TypeScript Mastery — 12 topics:

  • ▸Generics & Utility Types — Writing reusable, type-safe abstractions with conditional, mapped, and template literal types.
  • ▸Type Guards & Assertion Functions — Narrowing at runtime with is, asserts, and discriminated unions.
  • ▸Decorators — TC39 Stage 3 decorators vs legacy experimentalDecorators — when to use each.
  • ▸Zod — Runtime schema validation with full TypeScript inference.
  • ▸Declaration Files, Module Resolution & tsconfig — Understanding how TypeScript resolves modules and what each compiler option does.

Algorithms — 7 topics:

  • ▸Binary Search, Sorting, Graph Traversal (BFS/DFS), Trees, Dijkstra's Algorithm, Dynamic Programming, and Recursion — each with animated visualisations, complexity analysis, and interview-ready patterns.

Infrastructure as Code — 10 topics:

  • ▸Terraform — Modules, workspaces, state management, and drift detection.
  • ▸Bicep — Azure-native IaC with modules and Policy-as-Code.
  • ▸Pulumi — TypeScript and Python stacks for cloud infrastructure.
  • ▸Secrets Management — Vault, Azure Key Vault, and sealed secrets in GitOps workflows.

New Domains: Frontend React, Testing Engineering & Security Engineering

Expanding beyond AI and backend — this week we launched three new domains covering production-grade frontend engineering, testing strategy, and security.

Frontend React — 6 topics:

  • ▸Hooks — Custom hooks, rules, and common pitfalls in production apps.
  • ▸Server Components — React 19 async components, Suspense boundaries, streaming SSR.
  • ▸State Management — Zustand vs Jotai vs Redux Toolkit — decision guide with trade-offs.
  • ▸React Query — Server state, cache invalidation, optimistic updates.
  • ▸Next.js App Router & Fiber — Concurrent rendering internals and App Router migration strategies.

Testing Engineering — 8 topics:

  • ▸Unit, Integration, E2E, Contract, Snapshot testing — when to use each.
  • ▸TDD — Red-green-refactor rhythm with practical TypeScript examples.
  • ▸Test Architecture — Testing pyramid, diamond, and honeycomb patterns.
  • ▸Mocking Patterns — Spies, stubs, fakes, and MSW for HTTP mocking.

Security Engineering — 8 topics:

  • ▸OWASP Top 10 — Each vulnerability explained with mitigation code.
  • ▸JWT Auth, OAuth2 & OIDC — Token flows, refresh strategies, and common mistakes.
  • ▸Zero Trust Architecture — Identity-first security for cloud-native systems.
  • ▸Threat Modelling, SAST/DAST, Secrets Management — DevSecOps integration.

New Domains: Backend .NET, Cloud Azure & API Design

Three new domains this week aimed at engineers building production services on the Microsoft stack and designing robust APIs.

Backend .NET — 9 topics:

  • ▸Clean Architecture — Domain, Application, Infrastructure, Presentation layers in ASP.NET Core with dependency inversion.
  • ▸CQRS — Command/Query separation with MediatR, validation pipelines, and event sourcing.
  • ▸Minimal APIs — High-performance endpoints with built-in OpenAPI, filters, and typed route groups.
  • ▸Entity Framework Core — Query optimisation, change tracking, migration strategies.
  • ▸Dependency Injection, Async/Await, LINQ, Generics, Repository Pattern — Production patterns with gotchas explained.

Cloud Azure — 6 topics:

  • ▸Azure OpenAI — Deploying GPT-4o and embeddings on Azure, quota management, private endpoints.
  • ▸Azure Kubernetes Service (AKS) — Node pools, autoscaling, GitOps with Flux.
  • ▸Azure Functions — Durable Functions, bindings, cold start mitigation.
  • ▸App Service, Service Bus, Event Grid — Hosting, messaging, and event-driven architecture on Azure.

API Design — 8 topics:

  • ▸REST Principles, OpenAPI Spec, API Versioning — Pragmatic REST with contract-first design.
  • ▸GraphQL — Schema stitching, DataLoader, subscriptions.
  • ▸gRPC — Protobuf, streaming, and .NET integration.
  • ▸API Gateway & Rate Limiting — Kong, APIM, token-bucket patterns.
  • ▸WebSockets — Real-time bidirectional communication with backpressure.

New Domains: System Design, Databases & DevOps/CI-CD

This week we launched three foundational engineering domains — content that every engineer building at scale needs before designing AI systems.

System Design — 6 topics:

  • ▸Caching — L1/L2/L3 caching strategies, cache eviction policies, write-through vs write-behind.
  • ▸Rate Limiting — Token bucket, sliding window, fixed window — with distributed implementation guides.
  • ▸Load Balancing — Round-robin, least connections, consistent hashing for session stickiness.
  • ▸Message Queues — Kafka, RabbitMQ, Azure Service Bus patterns for async decoupling.
  • ▸Consistency & Sharding — CAP theorem, eventual consistency, horizontal partitioning strategies.

Databases — 12 topics:

  • ▸PostgreSQL & SQL Patterns — EXPLAIN ANALYSE, index strategies, CTEs, window functions.
  • ▸Vector Databases — pgvector, Pinecone, Weaviate — chunk storage, HNSW indexing, ANN search.
  • ▸Redis — Caching, pub/sub, Lua scripting, and Redis Streams.
  • ▸MongoDB, Cosmos DB — Document modelling, aggregation pipelines, multi-region write.
  • ▸Schema Design, Normalisation, Partitioning, Transactions — Practical patterns for production databases.

DevOps/CI-CD — 11 topics:

  • ▸GitHub Actions — Reusable workflows, matrix builds, composite actions, OIDC for Azure.
  • ▸Docker & Container Registry — Multi-stage builds, layer caching, image scanning.
  • ▸Kubernetes Deployments, Helm — Rolling updates, rollbacks, Helm chart best practices.
  • ▸Pipeline Design — GitOps, trunk-based development, feature flags in CI.
  • ▸Monitoring, Logging, Alerting, Service Mesh — Observability pipeline for production services.

New Domain: AI Observability — Monitor, Evaluate & Trust Your AI Systems

AI systems fail silently. This week we launched the AI Observability domain — 12 deep-dive topics on how to monitor, evaluate, and debug production LLM applications.

12 topics shipped:

  • ▸LLM Monitoring — Tracking latency, error rates, and cost per request in production.
  • ▸Token Tracking — Per-model cost attribution, budget alerts, and usage dashboards.
  • ▸Latency Analysis — Breaking down TTFT (Time to First Token), streaming overhead, queue wait.
  • ▸Online Evaluations — Real-time quality scoring on live traffic without ground truth labels.
  • ▸Offline Evaluations — Dataset-based regression testing with LLM-as-judge.
  • ▸Evaluation Frameworks — DeepEval, Ragas, PromptFoo — when to use each.
  • ▸LangSmith — Tracing LangChain pipelines, tagging datasets, running evals.
  • ▸Phoenix (Arize) — OpenInference tracing, embedding drift, cluster visualisation.
  • ▸OpenTelemetry for AI — Instrumentation standards for LLM spans and trace propagation.
  • ▸Input/Output Validation — Schema enforcement, guardrail policies, PII detection.
  • ▸Content Safety — Azure Content Safety, Llama Guard, custom classifiers.

Every topic includes a real implementation guide for observability pipelines.

New Domain: LLM Landscape — Model Comparisons, Benchmarks & Selection Guides

Choosing the right model is one of the highest-leverage decisions in AI engineering. This week we launched the LLM Landscape domain — 12 topics covering the model universe in depth.

12 topics shipped:

  • ▸GPT-4o — Architecture changes, vision capabilities, structured output JSON mode, pricing tiers.
  • ▸Claude (Anthropic) — Constitutional AI, extended context (200K), tool use, and safety evals.
  • ▸Gemini — Google's multimodal model family — 1.5 Pro, Flash, and Nano for on-device inference.
  • ▸Llama (Meta) — Open-weight models — fine-tuning, quantisation (GGUF/GPTQ), local serving.
  • ▸Self-Hosted Models — vLLM, Ollama, TGI — deployment, batching, and KV cache management.
  • ▸Edge Models — Phi-3, Mistral 7B, Gemma — running LLMs on-device and at the edge.
  • ▸API Providers — OpenAI, Anthropic, Groq, Together AI, Azure OpenAI — routing and fallback strategies.
  • ▸Context Windows — How 1M-token contexts change RAG, memory, and cost models.
  • ▸Benchmarks — MMLU, HumanEval, LMSYS Chatbot Arena — what they measure and their blind spots.
  • ▸Cost vs Performance — Token cost curves, latency/quality trade-offs, model tiering for multi-turn apps.
  • ▸Hallucination Testing — FactScore, HaluEval, custom probe suites for production validation.
  • ▸Evals — Building eval pipelines that actually predict production quality.

New Domain: Prompt Engineering — From Zero-Shot to Meta-Prompting

Prompt engineering is the most underrated engineering discipline in AI. This week we launched 12 production-focused topics covering the full spectrum from basic techniques to adversarial testing.

12 topics shipped:

  • ▸System Prompts — Persona, format, constraint, and chain directives. How instruction order affects model behaviour.
  • ▸Few-Shot Prompting — Selecting effective exemplars, formatting for consistency, dynamic shot selection.
  • ▸Zero-Shot Prompting — When zero-shot works and when it doesn't — with capability probing techniques.
  • ▸Chain-of-Thought (CoT) — Standard CoT, self-consistency, and process reward models.
  • ▸Tree of Thought (ToT) — Deliberate search over reasoning paths for complex problems.
  • ▸ReAct Prompting — Reason + Act loops for tool-augmented agents.
  • ▸Function Calling — Structured tool invocation with JSON schemas, parallel calls, and result handling.
  • ▸JSON Mode & Constrained Generation — Guaranteed structured output with Instructor and Outlines.
  • ▸Meta-Prompting — Prompts that generate or improve other prompts.
  • ▸Prompt Injection — Direct, indirect, and jailbreak attacks — detection and mitigation strategies.
  • ▸Red Teaming — Systematic adversarial testing of prompt-based systems.

All topics include templates, anti-patterns, and real production use cases.

New Domain: Agentic Systems — Build Autonomous AI Agents

Agents are where AI engineering gets hard. This week we launched the Agentic Systems domain — 12 in-depth topics covering everything from ReAct loops to multi-agent orchestration.

12 topics shipped:

  • ▸Agent Architecture — ReAct, Plan-and-Execute, Reflexion, and Self-Ask patterns compared.
  • ▸Tool Use — Tool schemas, result parsing, error recovery, and parallel tool execution.
  • ▸Tool Registry — Dynamic tool registration, capability discovery, and versioning.
  • ▸Planning Loops — HTN planning, goal decomposition, and plan validation strategies.
  • ▸Short-Term Memory — Conversation buffers, sliding windows, and summarisation memory.
  • ▸Long-Term Memory — Vector memory, episodic retrieval, and memory consolidation.
  • ▸Episodic Memory — Storing and retrieving past experiences to inform future decisions.
  • ▸Multi-Agent Orchestration — Supervisor, swarm, and hierarchical agent patterns.
  • ▸Agent Evaluation — Trajectory evaluation, success rate metrics, and agent benchmarks.
  • ▸LangGraph — Stateful multi-actor graphs, checkpointing, and human-in-the-loop flows.
  • ▸AutoGen — Microsoft's multi-agent conversation framework — group chats and code execution.
  • ▸MCP Integration — Model Context Protocol — giving agents access to tools, files, and data sources.

Each topic includes architecture diagrams, code examples, and production trade-offs.

New Domain: AI Engineering — Production RAG, Embeddings & LLM Systems

The first major content domain launches today. AI Engineering covers the core building blocks of production AI systems — from RAG pipeline design to token cost management.

15 topics shipped at launch:

  • ▸RAG (Retrieval-Augmented Generation) — Chunking, indexing, retrieval, re-ranking, and generation. Hybrid search with sparse + dense vectors.
  • ▸Vector Search — HNSW, IVF, PQ indexing. Similarity metrics: cosine, dot product, Euclidean.
  • ▸Embeddings — Embedding models (text-embedding-3, BGE, Cohere), dimensionality, fine-tuning.
  • ▸Multi-Agent Systems — Orchestrator-worker, supervisor, and mesh topologies.
  • ▸Agent Patterns — ReAct, Reflexion, Plan-and-Execute with LangChain and from-scratch examples.
  • ▸Agentic Workflows — Routing, parallelisation, human-in-the-loop, and subagent delegation.
  • ▸LLM Streaming — Server-Sent Events, token streaming, abort signals, partial JSON parsing.
  • ▸Token Economics — Context window costs, KV cache, prompt caching, and cost attribution.
  • ▸Fine-Tuning — LoRA, QLoRA, PEFT methods — when fine-tuning beats prompting.
  • ▸Model Serving — vLLM, Triton Inference Server, continuous batching, quantisation trade-offs.
  • ▸Guardrails — Input validation, output scanning, PII redaction, refusal detection.
  • ▸LLM Evaluation — RAGAS, G-Eval, LLM-as-judge, reference-free metrics.
  • ▸Semantic Kernel — Microsoft's SDK for AI orchestration in .NET and Python.
  • ▸MCP (Model Context Protocol) — Anthropic's open protocol for tool-augmented LLMs.
  • ▸Prompt Engineering — System prompts, few-shot, chain-of-thought, and structured output.

AI Wisdom Beta Launch — The Engineer's Knowledge Platform for AI

AI Wisdom is live. After months of building, we're opening the platform to the first cohort of engineers.

What AI Wisdom is:

AI Wisdom is a structured knowledge platform for engineers building production AI systems. Not tutorials. Not marketing fluff. Opinionated, production-focused guides written by an engineer who has shipped AI systems and made the mistakes so you don't have to.

Platform architecture at launch:

  • ▸Learn — Topic-based learning with See It (visualisation), Animate It (step-by-step flow), Build It (production code), and Quiz modes.
  • ▸Playground — Interactive AI engineering tools: RAG Tester, Prompt Lab, Token Counter, Embedding Explorer.
  • ▸Stories — Real-world architecture case studies with annotated technical decisions.
  • ▸Landscape — Curated guide to tools, frameworks, and providers in the AI engineering ecosystem.

Engineering stack:

  • ▸Next.js 15 App Router, TypeScript, Neon Postgres, Tailwind CSS.
  • ▸Content served from a Postgres database with full-text and semantic search.
  • ▸Deployed on Vercel with edge caching for sub-100ms TTFB.

What's coming next:

Weekly content drops across AI Engineering, Agentic Systems, Prompt Engineering, LLM Landscape, and System Design. Subscribe to get notified.