Module 1 · Fundamentals

What We Owe Machines

The hard problem in AI isn't making machines smarter, it's teaching them to handle problems without right answers. This lecture traces how we learned to teach machines at all, and where that project stands today.

Slides Lecture Notes Recording
Lecture notes by Riccardo Colletti

The Lifecycle of a Language Model

Understanding the full training pipeline from pretraining to deployment.

Slides Lecture Notes Recording

Module 2 · Post-Training

Post-Training Foundations

RLHF and Reward Learning

Alignment Methods & Model Behavior

Evals as Research

Module 3 · Reasoning & Agents

Search, Planning, Memory

Tool Use and Verification

Spring Break

No class this week. Enjoy your break!

Multi-Agent Systems

Module 4 · Product & Research

Product Design & Development Workshop

Product Workshop (Continued)

Guest Lecture

Guest Lecture

Demo Day 🎉

Date
Lecture
Technical Work
Materials
Assignments Due
Jan
26
Week 1 · Fundamentals
What We Owe Machines
Feb
2
Week 2 · Fundamentals
The Lifecycle of a Language Model
T2a: Build a Context-Aware Assistant
T2b: Inference Deep-Dive
Feb
9
Week 3 · Post-Training
Post-Training Foundations
T3a: Datagen Pipelines & Quality
T3b: SFT from Scratch
Feb
16
Week 4 · Post-Training
RLHF and Reward Learning
T4a: DPO in Practice
T4b: Toy Reward Modeling
Project 1 Released
Feb
23
Week 5 · Post-Training
Alignment Methods & Model Behavior
T5a: Refusal Classifier
T5b: Constitutional AI Loop
Final Project B Due
Mar
2
Week 6 · Post-Training
Evals as Research
T6a: Evaluation Design
T6b: Benchmark Hacking
Mar
9
Week 7 · Reasoning & Agents
Search, Planning, Memory
T7a: Planning Agent
T7b: AlphaEvolve
Project 1 Due
Mar
16
Week 8 · Reasoning & Agents
Tool Use and Verification
T8a: Coding Agent with Verification
T8b: Computer Use Lab
Project 2 Released
Mar
23
Spring Break
Mar
30
Week 9 · Reasoning & Agents
Multi-Agent Systems
T9a: Agent Telemetry
T9b: Multi-Agent Collaboration
Final Project C Due
Apr
6
Week 10 · Product
Product Design & Development Workshop
T10: Self-Play Experiment
Project 2 Due
Apr
13
Week 11 · Product
Product Design & Development Workshop (Ct'd)
Apr
20
Week 12 · Guest Lecture
TBA
Apr
27
Week 13 · Guest Lecture
TBA
May
4
RRR Week
Demo Day 🎉
Final Project Due
Final Project

Final Project

Design and build a product demo AI product prototype that demonstrates mastery of course concepts. Teams of 2-4 students will propose, build, and present a demo.

Team Formation Feb 9
Proposal Due Feb 23
WIP Check-in Mar 30
Demo Day May 4

The final project is your opportunity to explore a topic of your choice in depth. Projects should demonstrate technical sophistication and original thinking. Successful projects might included novel alignment techniques, agent architectures, evaluation frameworks, and creative applications of post-training methods.

Milestones

Feb 9
Checkpoint A: Team formation and initial idea submission
Feb 23
Checkpoint B: Detailed proposal with methodology and timeline
Mar 30
Checkpoint C: Work-in-progress presentation and feedback session
May 4
Final submission and Demo Day presentation

About the Course

This project-based course teaches you to design, build, post-train, and evaluate agentic AI systems. We connect core technical foundations (model architecture, retrieval-augmented generation, tool use, multi-agent coordination) with hands-on labs and a team capstone project.

You'll gain practical experience with model deployment frameworks, post-training pipelines (SFT, RL, context engineering), evaluation systems, and model-context protocols. The course culminates in building an end-to-end minimum viable AI product.

Emphasis is placed on engineering rigor, creative problem-solving, and deployment at scale.

Learning Outcomes

By the end of this course, you will be able to:

  • Design AI systems that are reliable, maintainable, and cost-effective
  • Debug and align models in production using modern evaluation and feedback frameworks
  • Post-train open-source LLMs using SFT, RL, and context engineering via TinkerAPI
  • Translate research into practice, whether founding an AI venture or shipping at scale
  • Navigate ethical considerations in deploying autonomous systems
  • Prototype and iterate on applications using modern LLM and agentic architectures

Who This Course Is For

Upper-division undergraduates and early graduate students who want to found or join early-stage AI startups, or work on applied ML engineering at major technology companies.

Expected background: Solid Python programming experience (CS 61A/88 + CS 61B or equivalent). Prior ML exposure recommended (CS 182, CS 189, CS 185, or Data 100) but not required. Strong motivation to apply AI research to real-world systems.

Course Structure

  • Module 1: Fundamentals Foundations of LLMs, architectures, inference
  • Module 2: Post-Training SFT, RLHF, alignment techniques, TinkerAPI
  • Module 3: Reasoning and Agents Chain-of-thought, tool use, multi-agent systems
  • Module 4: Product and Research Deployment, evaluation, scaling, ethics

Navigating the Course

We keep things simple: one channel for communication, one for submissions, and one site for all course content. Everything you need lives in one of these four places:

  • Communication: Slack (see welcome email for invite link)
  • Course website: posttraining.ai
  • Assignments: Gradescope (enroll with code 6X7K5K)
  • Lectures: In-person, recordings available on course site

Grade Breakdown

50% Final Project (semester-long, 2 checkpoints, Demo Day May 4th)
20% Project 1 & Project 2 (10% each)
20% Attendance (in-person required, 2 excused absences allowed)
10% Documentation (weekly progress sharing, starting week 2)

Assignments

Attendance (20%)

We believe you learn best by being in class and discussing with your peers. In-person attendance is required with 2 excused absences allowed.

Documentation (10%)

Each week you'll share something you learned: a blog post, tweet, TikTok, LinkedIn post, YouTube video. The format is yours; the goal is building in public.

Labs (ungraded)

Weekly technical work (T1–T10) released alongside lectures. These are hands-on labs that reinforce lecture topics and prepare you for the graded projects. Ungraded but strongly recommended.

Projects (graded)

Project 1 focuses on post-training pipelines. Project 2 focuses on agentic systems. Both are graded on completion and quality.

The Final Project is a semester-long endeavor where you ideate, design, prototype, and productionize an AI system. Two checkpoints throughout the semester, culminating in Demo Day on May 4th.

Acknowledgements

  • Cognition for providing compute credits
  • Thinking Machines for TinkerAPI access