FAQ

I am the creator of What’s AI, co-founder and CTO of Towards AI, co-author of Building LLMs for Production, and an AI engineering educator and speaker. Most of my work is about making the technical decisions behind useful AI systems clearer and more practical.

What’s AI is my public work across videos, articles, newsletters, guides, and conversations about AI engineering. I start from a new technique or announcement, then ask what it changes for someone who has to build, evaluate, or ship a system.

Start with the simplest solution that can work. Make the workflow clear, define what a good answer means, add evaluation and observability, then increase autonomy only when the problem requires it. Reliability is usually harder, and more important, than prompting.

I am doubling down on the What’s AI YouTube channel, building B2B AI training with Towards AI, improving the Towards AI Academy platform and courses, and working on the second edition of Building LLMs for Production. I also have a few other projects in progress that I am not ready to share yet.

Yes. Start with vocabulary and intuition, then learn enough Python to read documentation, call APIs, work with data, and debug what a coding agent generates. You do not need to master everything before building your first small project.

I would go in this order: intuition and vocabulary, Python if needed, one structured course, one framework’s documentation, two or three small projects, then evaluation, tracing, cost, and deployment. Move into RAG, fine-tuning, or agents when the problem gives you a reason.

Not for most AI engineering roles. Mathematics becomes important depending on the problem, but you can begin with intuition, code, data, and projects. Fundamental research roles usually require a different level of depth.

Choose a narrow problem with data you can use responsibly: a document assistant with citations, a support workflow with structured outputs, or a document extractor with validation. Add a small evaluation set and write down the failure cases.

For intuition, start with What’s AI, 3Blue1Brown, Andrej Karpathy, Yannic Kilcher, or AI Coffee Break with Letitia. For structure, use the Hugging Face LLM Course, the AI Engineering Roadmap, or a Towards AI program that matches your level. Pick one path, finish a project, then come back when you hit a real limitation.

Move from notebooks to systems. Build an end-to-end service, add structured outputs, a reproducible evaluation, traces, a cost and latency budget, and deploy it. Then compare a simple approach with a more complex one instead of reaching for the newest framework first.

Two to four serious projects beat a page of demos. Show the problem, baseline, architecture decisions, tradeoffs, failures, evaluation, and what you would do next. A strong README proves you understand the system, not only that you can generate code.

Practice with a small, intentionally vague problem. Clarify assumptions, create a baseline, define an output schema and evaluation set, document the tradeoffs, and ship something another person can run. Judgment and communication matter as much as the result.

First learn to frame the problem and evaluate a baseline. Prompting shapes behavior, RAG supplies dynamic knowledge, fine-tuning changes model behavior, and agents add autonomy. Choose based on an observed limitation, not popularity.

Follow a small set of trusted sources, read release notes for tools you really use, and test changes only when they connect to a real problem. Data, context, evaluation, observability, and tradeoffs age better than framework lists.

Towards AI is an AI deployment and education firm. It combines education, community, workforce training, and system building so lessons from production flow back into the courses and practices people learn from.

Both. Individuals use the courses, book, content, and community. Companies work with us to train their teams and deploy custom AI systems that move past the demo stage.

If you do not code, start with a Python or AI-for-work path. If you know Python and want to build a full product, choose Full Stack AI Engineering. If you already have the foundations and want autonomous systems, look at Agentic AI Engineering.

The book covers LLM applications beyond the demo: prompting, RAG, fine-tuning, agents, evaluation, optimization, and deployment. It works well as a practical reference you return to by problem rather than a book you memorize front to back.

The Learn AI Together community includes beginners, students, engineers, and people building products. It is the best place for a general learning question that may also help other members.

Yes. Towards AI works with technical teams, non-technical teams, and leadership on training, strategy, use-case selection, and deployment. Send the context, audience, desired outcome, timeline, and important constraints.

Yes. Topics usually include AI engineering, agents and workflows, RAG, evaluation, context, memory, and production systems. Include the event, date, audience, format, location, and desired outcome in your request.

Yes when the product is genuinely relevant to people learning or building with AI. Send the product, audience fit, campaign window, goal, and useful links. Requests without context or a clear fit are less likely to move forward.

Use the contact form and choose the most specific topic. It routes training, speaking, partnership, media, and collaboration requests correctly. For broad learning questions, use the community instead.