Questions about AI Wisdom

A reference and a set of learning paths for engineers who build AI features that have to work in production: RAG, agents, evals, observability and the cloud, API and data work around them. It focuses on the decisions and the failure modes, not on what a model can do in a demo.

Amit Upadhyay, a software engineer working on AI systems and architecture. The examples come from building and running these systems, and the views are his own, not an employer’s.

About the author →

Yes. Every article and every learning mode is free to read without an account. An account is only needed to comment, react to an article, or vote on whether a page helped.

No advertising cookies. The site uses the cookies sign-in needs and privacy-conscious analytics that count page views in aggregate, so it is clear which pages are useful.

Privacy policy →

If you are building with AI, start with the AI engineering path on the homepage: RAG, then agents, evals, observability and serving, in the order those problems show up. If you need a refresher on the foundations first, every domain on the Learn page works on its own.

The AI engineering path →All domains →

Every topic can be learned five ways. Read is the article. See is the architecture as diagrams. Animate steps through the flow one stage at a time. Test is a question set with explanations. Build is a guided implementation with code you can run.

No. Read is enough to understand a topic. The other modes are there for when you want to see it, check yourself, or build it. Many readers use Read and Build only.

Your learning passport (completed modes, XP and streaks) is stored locally in your browser. It does not follow you to another device, and clearing site data resets it.

Learning passport →

The certification page shows the path and tracks your readiness. The advanced credential is paused while the free foundations path matures, so there is nothing to pay for or sit yet.

Certification →

Every article shows when it was published and when it was last updated. Architecture guidance changes slowly. Model names, prices and API details change fast, so for those, check the updated date and the provider’s own pricing page before you rely on a number.

Azure OpenAI ships models on its own schedule, usually behind OpenAI’s own API. An Azure article names what is actually deployable on Azure today, which can be an earlier generation than the one an OpenAI article describes. Both are correct for their platform.

Leave a comment on the page (it needs a free account) or post in the Discord. Corrections are fixed in the article and the updated date moves, so the next reader gets the fix.

Discord ↗

Yes, for learning and as a starting point. They are written to teach one idea clearly, so review, test and secure them before anything reaches production. You can link to any page and quote short excerpts with attribution; republishing whole articles needs permission.

Terms →

The glossary defines technical terms (what an embedding or a reranker is). The changelog records what changed on the site and when. This FAQ answers questions about using the site itself: what it costs, where to start, how accurate it is.

Glossary →Changelog →

Subscribe with the bar at the bottom of any page for one email a week when a new article is published, or follow along on Discord, YouTube, X, Reddit or Dev.to.