lb.

Relaunching learnbackend.com: AI Engineering for Backend Developers

July 9, 2026 · 2 min read

I'm a backend engineer at a fintech. Over the last two years, my job quietly changed: the systems I ship increasingly have an LLM somewhere in the request path. Nobody sent a memo. The queue consumers, the Postgres schemas, the retry logic: all still there. But now there's also a prompt under version control, a vector index next to the search cluster, and a line item on the cloud bill that scales with tokens.

If that's happening to me, it's happening to you, or it will soon. As of mid-2026, demand for engineers who can build around models (retrieval, agents, evals, serving) is growing faster than any other engineering specialty. And almost none of it is machine learning. It's backend engineering with new failure modes.

The gap this site fills

Most AI education assumes the wrong audience. Beginner courses explain what an API is. Cohort courses cost $750–$1,500 and demand fixed evening hours. YouTube is a slot machine.

What I wanted, and couldn't find, was the equivalent of a good internal engineering doc: dense, text-first, written for someone who already ships production systems and just needs the LLM layer mapped onto what they know. Tokens explained as a resource limit, not magic. Model selection framed like database selection. Evals framed as your test suite, because that's what they are.

So I'm writing it.

The track

AI Engineering for Backend Developers is free, and it ships in order, one module at a time:

  1. LLM Foundations: a systems-person's mental model (live now)
  2. Working with LLM APIs: the model as a flaky upstream dependency
  3. Prompting & Structured Output: prompts as API contracts
  4. RAG & Embeddings: your search infrastructure, upgraded
  5. Tool Use & Agents: job orchestration with an unreliable worker
  6. Evals & Observability: the skill that separates demos from products
  7. Serving, Cost & Latency: LLM performance engineering
  8. Capstone Projects: three production-grade builds

Every lesson assumes you know Postgres, queues, and HTTP. That assumption is the whole point: it's what lets the lessons be short.

Building in public

I'm building this on nights and weekends, and I'll write honestly about how it goes: traffic, what works, what flops. If you want new modules (and the occasional production postmortem) in your inbox, there's a weekly newsletter. No spam, no daily drip.

See you in lesson one.