How others build roadmaps (research, Sept 2026)¶
Before designing this roadmap, I analysed 17 public roadmaps and study systems. Here's what they do well, where they fall short, and what this site borrowed.
Examples studied¶
| Roadmap | Shape | Borrowed | Weakness |
|---|---|---|---|
| Lamarana12200/ai-engineering-roadmap | 20 wks × 13 h, 7 phases, 8/2/2/1 h split | Phase gates, "Explain-it-Back Friday", an "ignorance log", one artifact per phase with a write-up | AI only |
| tal7aouy/LLM-Engineering | 24 wks, 6 phases + "2026 deep dives" | Big question bank; core path kept separate from deep dives | No tracking; starts at beginner level |
| codejunkie99/agent-roadmap-2026 | 17 wks, primary sources | AGENTS.md protocol: an agent personalises and maintains the plan | No quizzes |
| jugaldb — AI Engineering 101 | Daily 15-min reps + weekly build | Daily rep + weekly build rhythm | No tracking |
| Toleflaco/ai-engineer-roadmap-java | Spring AI phases 0–6 | ADRs as the progress log; Java/Spring AI path | No weekly granularity |
| JuanMelendres/cracking-code-interviews | MkDocs site, L1–L4 syllabus, 897 flashcards | Levelled L1–L4 questions, study packs, MkDocs + Pages | Very big, little AI |
| ai-infra-curriculum principal track | 12 modules, 1000-pt scoring | Soft skills as scored deliverables | About 680 h |
| donnemartin/system-design-primer | Topic index + Anki decks | Pairing design topics with spaced repetition | Dated for LLM-era systems |
| roadmap.sh AI Engineer · Architect | Node graphs | Checking for breadth gaps | No time dimension or depth |
| mlabonne/llm-course | Scientist/Engineer tracks + notebooks | Reference depth for fine-tuning | No schedule |
| Chip Huyen — AI Engineering | Book chapters | Backbone for the AI track | Book, not a plan |
| StaffEng | Archetypes, guides | Staff+ track framing | No plan |
Gaps no single roadmap covered, and how this one fills them¶
- AI engineering + Staff-level architecture + DSA + Java together. Those roadmaps cover one or two of these; this one runs them as parallel tracks with fixed weekly hours.
- Spaced repetition is rare. Here: L1–L4 hidden-answer questions everywhere, DSA re-solve schedule, and a Sunday review hour.
- Interview checkpoints are almost never scheduled. Here: 6 of them, scored on one rubric.
- Most roadmaps assume a beginner. This one starts with a week-0 baseline and a skip list.
- They go stale. This one has daily and weekly agents plus a freshness contract.
- Progress lives in someone's head. Here, GitHub issues feed
progress.yml, which feeds the dashboard streak and heatmap.