study skillsintermediate 8m2026-07-06

The 24-Topic Learning Series

A 24-topic, 36-session evergreen deep-dive on engineering, ML, and LLM systems — with spaced-repetition revisions built in.

What this is

A 24-topic learning series that goes deep on the engineering, ML, and LLM concepts that show up in real production systems — the ones you keep hitting in interviews, design reviews, and 2am incident calls.

It's designed around three constraints most study plans ignore:

You have a job. Sessions are 90 minutes, three times a week. Total load ≈ 4.5 hours/week live + ~1.5 hours async. Sustainable, not heroic.
You forget things. Every two new topics are followed by a revision session — recall prompts, blank-page brain-dumps, redo-the-drill. Spaced repetition is how understanding turns into retention.
You want depth, not surface. Each session is a real 90-minute deep-dive: math, code, diagrams, production numbers. Not "top 10 tips."

The shape

24 topic sessions — one deep dive per topic, 90 minutes each.
12 revision sessions — one every third slot, no new content, pure recall + drill.
36 sessions total over 12 weeks at 3 sessions/week.
Suggested cadence: Tue evening, Thu evening, Sat morning. Adjust to your life.

The rhythm

S01  →  S02  →  R01  →  S03  →  S04  →  R02  →  …  →  S23  →  S24  →  R12
 new     new    revise  new     new    revise         new     new    revise

After every two new topics, a revision session:

10 min flashcards — 10 recall prompts you should hit in under 30s each
25 min blank-page recall — write everything you remember, no notes, colour-code the gaps
25 min hands-on redo — reproduce one code snippet or diagram from scratch
20 min drill redo — one LeetCode or system-design prompt with a timer
10 min gap analysis — log what you missed, plan a re-read

The 5-track rotation

Sessions rotate across five tracks so you never do the same thing twice in a row:

LLM — attention, RAG, agents, serving, fine-tuning, eval, capstone
ML — gradient boosting, embeddings, MLOps, recommender systems
DE — Spark, Kafka, streaming, lakehouse, data modelling, platform ops
SYS — URL shortener, distributed systems, chat system, caching, API design
OOP — SOLID, design patterns, Python concurrency

The 24 topics

#	Topic	Track
01	Transformers Foundations	LLM
02	Spark Execution Model	DE
03	URL Shortener System Design	SYS
04	SOLID & Design Patterns	OOP
05	Gradient Boosting	ML
06	RAG End-to-End	LLM
07	Kafka Fundamentals	DE
08	Distributed Systems Core	SYS
09	Python Concurrency	OOP
10	Embeddings & Vector Spaces	ML
11	Lakehouse Architectures	DE
12	LLM Agents	LLM
13	Chat System Design	SYS
14	Streaming Fundamentals	DE
15	MLOps & Feature Stores	ML
16	LLM Evaluation	LLM
17	Caching & CDN	SYS
18	Recommender Systems	ML
19	Data Modelling	DE
20	LLM Serving	LLM
21	API Design	SYS
22	Fine-tuning LLMs	LLM
23	Data Platform Cost & Governance	DE
24	Production AI Agent Capstone	LLM

The 12 revisions

Each revises the two sessions immediately preceding it.

#	Covers	Link
R01	S01+S02	Revision 01
R02	S03+S04	Revision 02
R03	S05+S06	Revision 03
R04	S07+S08	Revision 04
R05	S09+S10	Revision 05
R06	S11+S12	Revision 06
R07	S13+S14	Revision 07
R08	S15+S16	Revision 08
R09	S17+S18	Revision 09
R10	S19+S20	Revision 10
R11	S21+S22	Revision 11
R12	S23+S24	Revision 12

How to use it

Sequentially. Each topic assumes the ones before it. The revisions only work if the topics before them are fresh.
Or as a reference. Every post stands alone. Land on any one cold and it should teach you something without needing the rest of the series.
With a real prep repo. Every session's assignment ends with a git commit — that's the artefact that proves you did it, not the tab you opened.

What you don't need

A subscription.
A GPU.
A prior credential.
Every reading link marked as required. Skim the pre-read, do the deep-dive, save the rest for the weekend.

Start here

Session 01 — Transformers Foundations →

← previous

Production AI Agent — Capstone (Prompting, Safety, Observability)