Resources and Recommended Reading

Jump by topic
Each section starts with “best starting links,” then deeper references.

This is our “trustworthy links” page: official docs, standards, benchmarks, and a few high-signal learning resources. If you’re new, start with Start Here.

How to use this page (fast)
  • If you want the “official truth”: go to Developer Docs and Standards & Policy.
  • If you want to compare models/tools: start with Benchmarks.
  • If you want plain-English learning: start with AI Basics and Safety.
  • Tip: use Ctrl+F (or Find on mobile) to search any word.
Your main categories (the 5 “core” labels)
Recommendation: keep these 5 as your “core labels” and treat any other labels as secondary tags.

AI Basics (Plain English)

Good mental models (quick reminders)
  • Models predict patterns — they don’t “know” things the way humans do.
  • Outputs can be fluent and wrong — verify when it matters.
  • Data + objective + constraints usually explain behavior more than “intelligence” does.

Benchmarks & Evaluation

Why this section matters

“This model is better” only makes sense if you know better at what. Benchmarks help — but they can mislead if treated like one universal scoreboard.

Tip: when you see a score, check the task, the data, and the rules.

Frameworks & Libraries (Core Tools)

If you build anything beyond “toy scripts,” you’ll eventually hit these libraries. This list focuses on stable foundations.

  • PyTorch docs — training, inference, tensors, tooling.
  • TensorFlow docs — tutorials, guides, deployment paths.
  • scikit-learn — classic ML baselines + evaluation helpers.
  • MLflow — experiment tracking and lifecycle basics.
Practical tip: baseline first (simple model + clean evaluation) before fancy architecture.

Governance, Standards & Policy

Standards won’t write your system for you — but they help you ask better questions and build repeatable process.

Risk management
Principles & policy
Management systems
  • ISO (incl. AI management system standards)
Practical tip: governance is mostly repeatable process (docs, checks, reviews), not perfect prediction.

MLOps & “Making It Real”

Many failures are production failures: monitoring, drift, data issues, unclear requirements, and weak feedback loops.

A simple production checklist mindset
  • What can go wrong? (inputs, prompts, limits, downtime)
  • How do we detect it? (logging, monitoring, alerts)
  • How do we respond? (fallbacks, retries, human review)

Privacy & Data Handling (Reader-Safe)

If you collect, store, or process user data, you need more than good intentions. Start with dependable references.

Practical tip: privacy is often “good engineering”: least privilege + clear retention + collect less.

Safety & Security (LLMs, Prompting, Risks)

LLM apps create new risk categories: prompt injection, data leakage, jailbreaks, and overly-trusting automation.

LLM app risks
Adversarial threats
Practical build docs
A calm “safe-by-default” mindset
  • Don’t blindly execute model output (especially in automation).
  • Separate user input from system instructions.
  • Log and review: prompts, tool calls, failures, edge cases.

Workflows & Automation (Practical Building)

For automations (especially content pipelines), reliable structure matters more than fancy tricks.

Automation rules we follow
  • Make steps idempotent (safe to re-run).
  • Validate inputs (HTML, links) before publishing.
  • Keep a human review gate for public-facing content when possible.

Suggest a resource

If you found a great official doc, standard, or benchmark link we should include, send it via our Contact page. We prefer stable sources (official docs, standards bodies, reputable orgs).