Skip to content
17 hands-on courses

Build skills in the right order.

Choose a course track or browse the full catalog. Every course combines visual explanations, interactive learning, coding practice, and knowledge checks.

17
Published courses
7
Completely free
10
Pro, Module 1 free
Course tracks

Pick a course track. Follow it in order.

Each track contains only courses available on LDS and recommends the order to take them.

Course catalog

Find the exact skill you need.

Courses are grouped by how learners use them, not by launch date. Search a topic or narrow the catalog by domain and access.

17 of 17 courses

Foundations

Start from first principles in SQL, Python, statistics, or language models.

4 courses

Analytics

Build the querying, data-wrangling, and experimentation skills used in decision teams.

4 courses

Statistics & ML

Move from uncertainty and probability to models you can train and evaluate.

2 courses

AI systems

Build, adapt, evaluate, and operate modern LLM applications end to end.

7 courses
Pro
AI systemsIntermediate to Advanced

Building RAG Systems & Vector Search

Build production retrieval-augmented generation from first principles, BM25, dense embeddings, hybrid search + RRF, chunking, MMR, cross-encoder reranking, HyDE, ANN indexes (IVF/HNSW), and rigorous evaluation (Recall@k, MRR, nDCG). Every concept measured before/after on one realistic corpus.

Helpwell running corpusEvery concept measured
8 modules~9 hours
Pro
AI systemsIntermediate to Advanced

Building AI Agents

Build AI agents from scratch, the agent loop, tool use & function calling, planning, reflection/self-correction, memory, multi-agent systems, evaluation, and guardrails. The model is a deterministic stand-in; every piece of agent machinery runs for real on the Helpwell support agent.

Helpwell support agent throughlineReal loop, faked LLM
8 modules~9 hours
Pro
AI systemsIntermediate to Advanced

AI Evals: Test, Measure & Ship LLM Apps

Evaluate LLM, RAG, and agent systems for real, error analysis & failure taxonomies, deterministic & semantic scorers, LLM-as-judge alignment (Cohen’s kappa), confidence intervals & power, McNemar regression tests, and a CI gate. Model outputs and judge verdicts are baked; every scorer, statistic, and gate runs for real on the Helpwell support assistant.

Helpwell eval suite throughlineFaked outputs, real machinery
8 modules~9 hours
Free
AI systemsBeginner to Advanced

Build a Tiny LLM: From Tokens to Text

Build a working language model from scratch: tokenizer (BPE), embeddings and position, attention, the transformer block, decoding, training, and interpretability. Every concept runs live in your browser on StoryByte, a real ~1M-parameter GPT trained on TinyStories; watch it write an original story one real token at a time.

StoryByte: a real 1M-param GPTBuild it from scratch in NumPy
8 modules6h 45m
Pro
AI systemsIntermediate to Advanced

Fine-Tuning LLMs: Make the Model Yours

The post-training sequel to Build a Tiny LLM. Take StoryByte, the real ~1.09M-parameter GPT, and make it follow requests: training data design, SFT with loss masking, LoRA/QLoRA, DPO preference tuning, knowledge distillation, and int8 quantization, ending with the playbook for real 7B+ models. Every number measured; every training run reproducible on a laptop CPU via the companion repo.

Real 1.09M-param modelSFT · LoRA · DPO · distillation
8 modules~7 hours
Pro
AI systemsBeginner to Advanced

Context Engineering & AI Workflows

Build one Helpwell support workflow from a naive prompt to a measured system: context windows, prompt contracts, structured outputs, budgeted packing, grounding, injection defense, routing, human review, and regression gates.

21 interactive labs42 runnable coding practices
7 modules5h 55m
Pro
AI systemsIntermediate to Advanced

Production LLM Systems

Turn one working LLM call into a service people can depend on. Follow a support ticket from request to release while you add service promises, traces, traffic controls, cost routing, failure protection, tool permissions, incident response, and rollback.

One service improved across 8 modules16 interactive operations labs
8 modules8h 30m