Reference & tools · 14 / 15
The short watch list and course plan
Use the resource that unblocks the next project step. My default split is roughly 60% building, 25% focused study and 15% writing, feedback and job preparation. The percentages are a planning suggestion; beginners will need more study time at first.
Watch in this order
CS50P: Functions, Variables ↗
Beginners first; engineers may skip after the diagnostic.
After watching: write and modify a small function without copying the lecture.CS50P: Unit Tests ↗
Beginners after Python basics; vibe coders early in the bridge.
After watching: write one test that catches a deliberately introduced bug.Full Stack LLM Bootcamp: LLM Foundations ↗
All routes before the first model feature.
After watching: explain tokens, context and inference in your own words.Full Stack LLM Bootcamp: Augmented Language Models ↗
After you have made and tested a direct model call.
After watching: draw your retrieval pipeline and identify where a wrong answer can originate.Full Stack LLM Bootcamp: LLMOps ↗
When your application works locally.
After watching: define your quality checks, deployment plan and failure recovery.The three Full Stack lectures are from 2023. I selected them for the concepts; use maintained SDK and hosting documentation for code, model names and prices. The links open official pages with their video players. New recommended videos are online links, not added offline downloads.
The course sequence
- Foundation, if needed: CS50: Introduction to Programming with Python ↗. Work the exercises. Use selected CS50 SQL ↗ units when you start storing data.
- Core application work: this guide’s build exercises plus FastAPI’s current tutorial ↗. The optional DeepLearning.AI systems lab ↗ is an alternative if you prefer a guided notebook.
- RAG, if you need more practice: DeepLearning.AI: Building and Evaluating Advanced RAG ↗. Use it after you have a baseline and evaluation questions.
- Tools and agents: Hugging Face: AI Agents Course ↗ Unit 1, then one framework only if the project needs it.
- ML depth after the first app: Google: Machine Learning Crash Course ↗ for classification, data, overfitting and evaluation concepts. Check the prerequisites ↗; add the full course for ML-heavy roles.
- Model training extension: Hugging Face: LLM Course ↗ after Python and introductory deep learning. It is not the starting course for this applied path.
Cost and access
CS50 OpenCourseWare, the Full Stack lectures, Hugging Face courses and the linked official docs provide a free study route. API calls, hosting and optional certificates may cost money. The two DeepLearning.AI listings were found in the official indexed catalog, but direct access was blocked during research; confirm their current enrollment terms before paying. They are optional, and the core plan does not depend on them.
What I would postpone
Training a foundation model, Kubernetes, several agent frameworks, advanced fine-tuning, and a long certificate sequence. Add each only when a target role or a measured project problem calls for it. Keep learning basic statistics and data judgment alongside the application work; postpone breadth, not understanding.