PrerequisiteEst. 0.5–1 hours
GitHub Setup
Set up your GitHub account, create your course repo, install Git, and make your first commit.
Start this weekCore deliverable
A working GitHub repo with your name in the README.
AI Learning Program
A hands-on course from neural network basics through building and shipping your own AI agents. Watch the videos, run the code, fill in your reflections, and commit your work to GitHub each week.
Open the main documents directly in the browser.
Send meaningful feedback about the course or the product via Telegram. The best feedback format is:
1. What I was trying to do
2. What felt confusing, slow, or annoying
3. What change would help
4. Importance: low, medium, or high
The baseline structure stays stable, but OnPoint watches for new source material. When tracked sources change, the system refreshes the live syllabus supplement and drafts an announcement for chat.
Every week ends in code, notes, and one improvement idea.
Tip: You can speed up videos to 1.25x or 1.5x if the pace feels slow. Just don't go so fast that you're hearing words without absorbing ideas. If you need to rewind, you're going too fast.
Set up your GitHub account, create your course repo, install Git, and make your first commit.
Start this weekCore deliverable
A working GitHub repo with your name in the README.
Start with a single neuron so weights, bias, and activation all feel concrete.
Start this weekCore deliverable
Run the neuron script, modify it, and explain what changed.
Required video
Files
Watch a simple model improve by following gradients downhill.
Start this weekCore deliverable
Generate a loss curve and compare learning-rate choices.
Required video
Files
Use Karpathy’s tiny autodiff engine to understand backprop instead of memorizing it.
Start this weekCore deliverable
Complete a guided micrograd modification and trace one backward pass.
Required video
Files
Build a tiny character model and see how sampling creates language-like output.
Start this weekCore deliverable
Generate multiple samples and compare temperature or dataset changes.
Required video
Turn attention from an abstract buzzword into a simple weighted lookup calculation.
Start this weekCore deliverable
Run the notebook and explain queries, keys, values, and weights.
Required video
Learn the GPT workflow without pretending a laptop should do giant-model training.
Start this weekCore deliverable
Prepare data, log experiments, and protect the CPU budget.
Required video
Rebuild familiar concepts with tensors, autograd, and a real training loop.
Start this weekCore deliverable
Train a small classifier and compare at least one hyperparameter change.
Files
Use a real API safely with local secrets instead of hardcoded keys.
Start this weekCore deliverable
Run a CLI chatbot and experiment with prompt design.
Required video
Retrieve from local notes before answering so the system has something grounded to work with.
Start this weekCore deliverable
Ask questions, inspect retrieved notes, and improve the sample knowledge base.
Required videos
Ship something real with good scope discipline and explain your decisions.
Start this weekCore deliverable
Customize the template, build a small app, and demo it.
Required video
Module 2
Build AI agents from scratch, deploy on OpenClaw, connect to Telegram, and ship an auto-research agent. Requires an Anthropic API key and a DigitalOcean account (parent credit card needed).
An agent is an LLM that can decide what to do next. Learn the ReAct pattern and build a simple agent loop.
Start this weekCore deliverable
Run the ReAct loop, ask it 3 questions, and paste the full reasoning trace into the reflection.
Required video
Give an LLM access to tools and watch it decide when to use them. This is the core pattern behind every useful agent.
Start this weekCore deliverable
Run the agent with all 3 tools, show it answering a question that requires multiple tool calls.
Required video
Real tasks require planning, executing, checking, and correcting. Build an agent that breaks problems into steps.
Start this weekCore deliverable
Run the agent on 2 complex questions, show the plan it generated and how it self-corrected.
Required video
Agents forget everything between calls. Add conversation memory and a knowledge store to make them useful.
Start this weekCore deliverable
Have a multi-turn conversation and save/retrieve notes. Show the agent remembering context.
Required video
You've built agents from scratch. Now use a real framework — deploy OpenClaw and write your first skill.
Start this weekCore deliverable
A working OpenClaw workspace with one custom skill. Screenshot of a conversation using the skill.
Required video
Connect your agent to Telegram so you can talk to it from your phone. Write a second skill that does something useful.
Start this weekCore deliverable
Send a message to your agent on Telegram, get a response using your custom skill. Screenshot the conversation.
Required video
Put it all together. Build an OpenClaw skill that searches the web, reads sources, and writes a cited research report.
Start this weekCore deliverable
Run the research agent on 2 topics. Commit the generated reports to your GitHub repo.
Required video
Ship an agent that solves a real problem for you. Two custom skills, connected to Telegram, something you actually use.
Start this weekCore deliverable
Working agent on Telegram with 2+ skills. README explaining what it does and why. Demo screenshots.
Required video