The Practitioner's LLM Curriculum
Self-paced · Free · Open · 2026 edition

Twelve weeks from knowing about LLMs to shipping them in production.

A practitioner-direct curriculum for engineers who shipped ML before but watched the LLM stack move past. Modern architectures, context engineering, RAG, agents, evaluation, inference cost, reliability, production patterns. Twenty-six interactive widgets that make the math visceral. No signup, no upsell.

12
weeks
26
interactive widgets
~40k
words of content
$0
cost · no signup
What this is

Built like a working engineer's notebook, not a textbook.

Every page is written like a senior engineer telling you what they wish they'd known six months ago. Concrete decisions, the math behind them, and the tradeoffs nobody publishes. No marketing copy. No "exciting world of AI." Just the engineering layer.

01

Practitioner-direct

Opinions stated as opinions. Numbers shown with their sources. Tradeoffs spelled out. The voice you'd want from a senior engineer who's shipped this stuff and is now teaching you what they wish they'd known.

02

Interactive throughout

Twenty-six widgets — calculators, simulators, decision matrices, trace visualizers, cost explorers — embedded inline. Slide a parameter; see the math change. Reading without doing builds fluent commentators, not practitioners.

03

Ships you to production

The curriculum ends with deployment patterns, on-call playbooks, cost governance, and senior interview prep. Capability is week 1-11; week 12 is the engineering discipline that keeps capability shipped.

The curriculum

Twelve weeks. The arc from attention to production reliability.

Each week is roughly 8–10 hours of work: reading, widgets, build exercises, optional deep reads. Designed in order — early weeks set vocabulary and mental models that later weeks rely on. You can skip ahead if a topic is already familiar; most weeks stand alone for the practitioner already at their level.

Week 01 10 hours · 4 widgets

Modern LLM architecture

Tokenization, attention variants in 2026, KV cache, sampling. The mental models everything else builds on.

Read this week
Week 02 9 hours · 2 widgets

Context engineering

Long-context economics, prompt caching, position effects, the lost-in-the-middle problem. Where the practitioner work happens.

Read this week
Week 03 8 hours · 2 widgets

Fine-tuning

When (and when not) to fine-tune. LoRA, QLoRA, full SFT. The decision matrix and the parameter math.

Read this week
Week 04 9 hours · 2 widgets

RL for LLMs

DPO, RLHF, IPO, KTO. How preference signals shape models. Why reward hacking is fundamental, not a bug.

Read this week
Week 05 9 hours · 2 widgets

RAG basics

Embeddings, vector search, chunking. The geometric intuition for why retrieval works (and why it breaks).

Read this week
Week 06 9 hours · 2 widgets

Advanced RAG

Hybrid search, rerankers, query rewriting. The 2026 production-RAG architecture and what makes it work.

Read this week
Week 07 9 hours · 2 widgets

Agents and tool use

How agent loops work, where they break, the discipline that keeps them bounded. Three traces, six tools, thirty cells of judgment.

Read this week
Week 08 9 hours · 2 widgets

Evaluation discipline

Eval is the single highest-leverage practice. Golden sets, LLM judges, judge-vs-human agreement, eval-in-prod.

Read this week
Week 09 9 hours · 2 widgets

Inference optimization

Cost arithmetic that works. Batching, speculative decoding, model routing, KV cache reuse. The math behind a 10× cost difference.

Read this week
Week 10 9 hours · 2 widgets

Multimodal & code agents

Vision pricing, multimodal token economics, code-agent traces. Where the next product wave is built.

Read this week
Week 11 9 hours · 2 widgets

Reliability & verification

Calibration, verifier ensembles, citation entailment, formal grounding. The reliability stack that holds up under load.

Read this week
Week 12 9 hours · 2 widgets

Production & interviews

Deployment patterns for non-deterministic systems, on-call playbook, cost governance, the 7-pillar interview frame.

Read this week
How a week works

Same shape every time, so the structure disappears.

After two or three weeks, you stop noticing the format and just absorb the content. Each week opens with a stated goal, builds across numbered sections with embedded widgets and quick checks, and closes with concrete build-this-week exercises and interview-prompt questions.

01

Goal & stakes

What you'll be able to do at the end. The honest answer to "why this week."

02

Numbered sections

Concrete, actionable angles on the topic. No filler, no recap of what you already know.

03

Inline widgets

Calculators, visualizers, simulators where the math becomes visceral.

04

QuickChecks

Small multiple-choice questions inline to verify the concept landed.

05

Build this week

Four to five concrete exercises. Pick at least two; do them this week, not someday.

06

Read this

Curated papers and reference posts. Foundational, not exhaustive.

07

Interview prompts

Ten questions an interviewer might ask. Practice your verbal answer.

08

What done looks like

Bullet list of capabilities you should have by week's end. Self-assessment.

Sample · Week 9 · Section 2

Cost arithmetic that survives the conversation with finance.

Most teams know their per-token pricing. Few can produce a defensible monthly forecast. The gap is arithmetic discipline — combining input/output token counts, request volume, model tier, and cache-hit rates into a number you'd stake your headcount on.

The right calculation is per-call cost × request volume × hours × days, with separate input and output token streams. Here's the calculator we'll use for the rest of this section:

⟨ inference-cost-calculator widget ⟩
interactive · 6 models · per-call and yearly totals

Slide the request rate to 50 RPS; watch Sonnet at default settings produce $237k/year. Switch to Haiku: $58k/year. The 4× ratio is the intuition you want before you walk into the architecture meeting.

QuickCheck
Your default model is Sonnet at $3/$15 per Mtok. You serve 10 RPS, average 4000 input + 200 output tokens. What's your monthly run rate?

The trap most teams fall into is reasoning about per-call cost in cents and never multiplying by the rate. Per-call cost looks fine; monthly burn looks like a year of headcount.

The widgets

Real interactive tools, not animations.

Every widget is single-file standalone HTML with the same design system. They're built to teach a concept by letting you slide parameters and see the consequence — cost numbers, latency curves, calibration arcs, agent traces. Click any to launch. They open in their own page; come back when you're done.

and twenty more across the curriculum
Who this is for

If you've shipped ML before but watched the LLM stack move past you.

The curriculum assumes basic ML literacy: you know what gradient descent is, you've read about transformers, you're comfortable reading code. Beyond that, no assumed background. The ramp is steep but the entry point is reasonable.

Built for you if

  • You're an engineer with classical ML or deep-learning background catching up on LLMs in 2026
  • You've built features with the OpenAI/Anthropic API but haven't gone deeper
  • You're interviewing for senior LLM-engineering roles and want to structure your study
  • You're already shipping LLMs and want sharper tools for the production layer
  • You want practitioner depth without the marketing fluff or paywalled gates

Probably wrong fit if

  • You're brand new to programming or have no ML background — start with a Python and basic ML tutorial first
  • You want a research-track curriculum focused on novel architectures or theoretical work
  • You want a guided cohort with deadlines, accountability, and grading
  • You only want code snippets — half the value here is reasoning prose around the code
Pricing model

Free to read. Forever.

The reading layer of this site is free and will stay free. No signup, no email gate, no "premium tier" of locked content. Anthropic, OpenAI, and Google's docs are free too — knowledge that gates itself doesn't compound.

FAQ

Common questions, direct answers.

How long does this actually take?
8-10 hours per week of focused work. If you do every build exercise and read every linked paper, more like 12-15. If you skim and skip the build exercises, 4-5. The honest target: complete in three months at one week per week, or two months at one-and-a-half per week. Faster than that and the practice doesn't compound.
Do I need to do the weeks in order?
Recommended for weeks 1-6; later weeks stand alone. Week 1 establishes vocabulary used everywhere later. Weeks 5-6 (RAG) build on each other. Weeks 8 (eval) and 11 (reliability) reference 7 (agents) and 10 (multimodal). If you're already past a topic, skip it; don't waste your time.
Why these specific 12 weeks and not, say, 24?
12 weeks is what fits a quarter or a focused interview prep cycle. Going longer dilutes attention; going shorter skips production-engineering material that's the hardest to learn outside a job. The arc — capability in weeks 1-11, production discipline in week 12 — is the whole curriculum compressed.
Will this content go stale?
Yes, partially. The model lineup, pricing, and benchmark comparisons need quarterly refresh. The conceptual material (attention, RAG, agents, eval, reliability) is much more stable. The site shows a "last reviewed" date on each week and will be refreshed on a stated cadence.
Can I use the content in my own teaching or company training?
Yes, with attribution. The widgets are MIT-licensed (see GitHub link in footer); the prose is Creative Commons BY-SA 4.0. Use it freely for non-commercial training; for commercial use, link back and credit.
Who built this?
Venkatesh Baglodi — Ph.D. in Computer Science, with published research in NLP and computer vision. I have built and led AI platform work in fintech, healthcare and telecommunications, and previously led science-engineering for Alexa Ranking & Arbitration at Amazon. Most recently I architected a patent-pending agentic risk-assessment system built on LLMs with retrieval-augmented generation and human-in-the-loop review. The about page has the longer version. The curriculum is designed to be replicable — every fact has a source, every widget has its math documented in the footer, every claim is either stated as opinion or backed by a reference.