A curated, opinionated path through the courses, tools, and references worth your time.
Not a link dump. Every entry answers three questions: what is it, who is it for, and how long will it actually take. Anything I haven't personally worked through is marked as such.
Entries live in resources.yml.
The README is generated — see CONTRIBUTING.md.
- Start here — 3 entries
- Courses — 8 entries
- Tools — 8 entries
- References — 4 entries
- How to read this
If you're new and want one thing to open first, open the first one.
Microsoft · free · beginner · ~20 hours · 21 lessons, Python + TypeScript
The best free on-ramp to building with LLMs. Lessons split cleanly into "Learn" (concepts) and "Build" (code you run). Needs an Azure OpenAI, GitHub Models, or OpenAI key — GitHub Models is the free path.
completed · genai llm prompt-engineering rag
Andrej Karpathy · free · intermediate · ~25 hours · Video lectures + notebooks
Builds backprop, then a transformer, from scratch in plain Python. The single best way to stop treating models as magic. Slow down and type the code rather than watching it.
completed · deep-learning transformers fundamentals
fast.ai · free · beginner · ~40 hours · Video course + fastbook
Top-down: you train a working model in lesson one and learn the theory afterwards. The opposite order to Karpathy, and a good complement to it.
partly done · deep-learning computer-vision nlp
Longer structured programmes. Costs and time estimates are the realistic ones, not the marketing ones.
Microsoft · free · beginner · ~24 hours · 26 lessons, Python + R
Classical ML, not deep learning. Regression, clustering, NLP, time series.
reviewed · machine-learning python r
Microsoft · free · beginner · ~20 hours · 20 lessons
Ethics, data prep, visualisation, lifecycle. Good for non-CS backgrounds.
reviewed · data-science visualization
Microsoft · free · beginner · ~24 hours · 12 weeks, 24 lessons
Broader and more theoretical than the GenAI course. Symbolic AI through neural networks.
reviewed · ai deep-learning fundamentals
Microsoft · free · intermediate · ~12 hours · Lessons + code
Agent design patterns, tool use, planning. Do the GenAI course first.
in progress · agents llm
Microsoft · free · intermediate · ~8 hours · Lessons + code
Model Context Protocol is becoming the standard way tools attach to models. Worth understanding before you write another bespoke tool schema.
not yet reviewed · mcp agents tooling
DataTalks.Club · free · beginner · ~4 months · Cohort course + projects
Deployment-focused. You ship models, not just train them. Runs as a cohort with deadlines.
reviewed · machine-learning mlops deployment
DataTalks.Club · free · intermediate · ~4 months · Cohort course + projects
dbt, Spark, Kafka, orchestration, warehousing. The gap most data scientists have.
not yet reviewed · data-engineering spark dbt
OSSU · free · beginner · 2+ years · Full curriculum
A complete degree-equivalent path. Realistically nobody finishes it, but it's an excellent map for finding the gaps in your own knowledge.
reference · curriculum data-science mathematics
Free, open source, and worth the install. Each line says what it replaces.
Ollama · free · beginner · 5 minutes · CLI
Run open language models locally. One command to install, one to pull a model.
daily driver · llm local-ai
vLLM project · free · advanced · ~1 hour · Python library / server
Production inference. Continuous batching and an OpenAI-compatible endpoint, which matters more than the throughput — swapping providers becomes a base-URL change.
reviewed · llm serving gpu
DuckDB Labs · free · beginner · 15 minutes · Embedded database
Query millions of rows straight from CSV or Parquet with no server. Benchmarks in my duckdb-vs-pandas repo if you want numbers.
daily driver · analytics sql data
OpenAI · free · beginner · 20 minutes · Python library
Speech to text, offline, 90+ languages. Use faster-whisper for real workloads.
daily driver · speech transcription
LangChain · free · intermediate · ~3 hours · Python library
Agent loops as state machines, with checkpointing and interrupt points. The distinction that matters: a chain runs steps you defined, an agent runs a loop where the model decides.
in progress · agents orchestration
Langfuse · free · intermediate · ~1 hour · Self-hosted service
Token cost per run, latency per node, prompt versioning. Grafana understands CPU; it does not understand tokens. You need both.
reviewed · observability evals llmops
Supabase · free tier · beginner · ~1 hour · Hosted or self-hosted
Postgres with auth, storage, realtime and pgvector. One dependency instead of four.
reviewed · database backend pgvector
IBM · free · intermediate · ~1 hour · Python library
Document parsing that preserves table structure. Mangled tables are the most common silent cause of confidently wrong RAG answers.
not yet reviewed · rag parsing documents
Things I reopen rather than read once.
Howard & Gugger · free · beginner · reference · Jupyter notebooks
Full text of the O'Reilly book as runnable notebooks.
reference · deep-learning book
Andrej Karpathy · free · intermediate · reference · Course repo
Build a storyteller LLM from scratch. Still evolving — check activity before committing.
not yet reviewed · llm from-scratch
Donne Martin · free · intermediate · reference · Markdown + flashcards
Not AI-specific, but the vocabulary you need the moment a model goes to production.
reference · system-design interviews
Hugging Face · free · intermediate · ~10 hours · Notebooks
Aligning and fine-tuning small models on modest hardware.
not yet reviewed · fine-tuning small-models
Status tells you how much weight to give my opinion:
| Status | Means |
|---|---|
daily driver |
I use this regularly |
completed |
I finished it |
in progress |
I'm working through it now |
partly done |
I did some of it |
reviewed |
I've used or read enough to judge it |
reference |
I reopen it rather than read it through |
not yet reviewed |
On my list — included for completeness, not endorsed |
Time estimates are realistic, not promotional. If a course says "6 hours" and actually takes 20 once you run the code, the number here is 20.
Most common tags: llm (5), deep-learning (4), agents (3), rag (2), fundamentals (2), machine-learning (2), data-science (2), genai (1)
Edit resources.yml, not this file. See CONTRIBUTING.md.
Links are checked weekly by CI. If one rots, an issue opens automatically.
MIT. Curation is opinion, not endorsement — check licences on the linked projects themselves.
Maintained by Ishita Sharma.