Free · text-first · no signup
The Future is Still Human.
Ship LLM systems like you ship backend systems.
Production AI engineering for backend developers. A free, text-first track: RAG, agents, evals, serving, cost and latency, in the language of APIs, databases, and queues.
~91 min live · free · no account needed
Assumes you know backend
No 'what is an API' filler. Every lesson maps LLM concepts onto what you already run: Postgres, queues, caching, distributed systems.
Production-first, not demo-first
Anyone can ship an LLM demo. This track is the other 90%: evals, observability, retries, guardrails, cost and latency engineering.
Text and code, no videos
Dense, skimmable lessons with runnable code. Read one over coffee, apply it at work the same day.
Lessons that read like good internal docs.
Dense prose plus runnable code, with a table of contents, progress that saves in your browser, and a backend analogy for every new concept.
- ⧉ Runnable code with one-click copy
- ▤ TOC, reading time, and progress (no account)
- ⇄ A backend analogy for every LLM concept
AI Engineering for Backend Developers / LLM Foundations
Module 1 · Lesson 2 of 6
Tokens, Context Windows, and Your New Resource Limits
8 min read
Every resource limit you've ever tuned (payload caps, connection pools, RAM) has an LLM equivalent, and they're all denominated in the same unit: the token…
import { encode } from "js-tiktoken";The road from tokens to production.
2 of 8 modules live. They ship in order, roughly one per week. Progress is saved in your browser.
- 01LLM FoundationsA working mental model of LLMs for people who think in systems: tokens, context windows, model selection, non-determinism, and the anatomy of the production AI stack.6 lessons
~47 min - 02Working with LLM APIsThe LLM API as a backend dependency: chat completions, streaming, retries and timeouts, rate limits, provider abstraction, and treating model calls like any other flaky upstream service.6 lessons
~44 min - 03Prompting & Structured OutputPrompts as API contracts: system prompt design, few-shot patterns, JSON schema outputs, tool/function definitions, and validating model output like untrusted user input.coming soon
- 04RAG & EmbeddingsRetrieval-augmented generation end to end: embeddings, chunking strategies, vector databases vs pgvector, hybrid search, reranking, and the architecture of a production RAG service.coming soon
- 05Tool Use & AgentsFrom single tool calls to agent loops: designing tools, orchestration patterns, state and memory, guardrails, human-in-the-loop, and when an agent is the wrong answer.coming soon
- 06Evals & ObservabilityThe skill that separates demos from products: building an eval harness, golden datasets, LLM-as-judge, regression testing prompts, tracing, and monitoring LLM systems in production.coming soon
- 07Serving, Cost & LatencyLLM performance engineering: token economics, caching strategies, model routing, batching, self-hosting vs API trade-offs, and keeping p95 latency and unit costs under control.coming soon
- 08Capstone ProjectsThree production-grade builds that tie the track together: a RAG service with hybrid search, an eval harness for a real prompt, and an observable agent with guardrails.coming soon
Why this exists
AI can write the code. Someone still has to ship the system: own the retries, the evals, the latency budget, the bill. This track exists to keep backend engineers in that seat.
— written by a backend engineer at a fintech, shipping LLM systems in production. Lessons come from building the real thing.
Questions, answered.
Is it really free?
Yes. The full track is free to read, no account, no trial, no unlock. If a paid deep-dive product ever exists, it will be separate and clearly labeled. The track itself stays free.
Do I need an ML background?
No. If you can build an API against Postgres and a queue, you have every prerequisite. The track deliberately teaches the engineering layer around models, not training them.
Who is this for?
Working backend engineers who are being asked to ship LLM features and want to do it with the same rigor they apply to the rest of their stack. It is not for absolute programming beginners.
Why text instead of videos?
Text is denser, skimmable, searchable, and copy-pasteable. A 43-minute read carries more than three hours of video, and you can read it in the gaps of a workday.
How often do new lessons ship?
Roughly one module per week until all eight are live. The newsletter announces each one; the RSS feed carries them too.
Learn to ship the systems everyone else only prompts.
or get one lesson a week by email: