Self-Paced Study Plan 2027

Overview

This plan builds on 2026’s foundations. The central focus is a deep, structured study of AI/ML — from fundamentals to production systems — complemented by continued philosophy reading and privacy engineering.

Study Streams:

1. Artificial Intelligence & Machine Learning (Deep Dive)
2. Philosophy & Humanities (continuation)
3. Privacy Engineering (continuation)

Study Stream 1: Artificial Intelligence & Machine Learning

Goal

Build a rigorous, bottom-up understanding of modern AI — from neural network fundamentals through LLMs, generative models, RL, and production ML systems.

Curriculum

Eight courses in progressive order, from foundations to specialization. Each course should be completed before moving to the next, with a hands-on project at the end of each block.

Block 1: Foundations (Q1 — Jan to Mar)

Course 1 — MIT 6.S191: Introduction to Deep Learning

• Provider: MIT
• Focus: Neural nets, CNNs, transformers, generative models — deep learning from zero
• Format: 10 lectures (YouTube playlist)
• Complete before moving to Karpathy

Course 2 — Zero to Hero

• Provider: Andrej Karpathy
• Focus: Build a GPT from scratch — backprop, tokenization, attention mechanisms
• Link: karpathy.ai/zero-to-hero.html
• Most hands-on course in the stack — take notes, implement everything

Milestone: Implement a small language model from scratch by end of March

Block 2: Language Models & Generative AI (Q2 — Apr to Jun)

Course 3 — CS336: Language Models from Scratch

• Provider: Stanford
• Focus: Data preprocessing, scaling laws, evals, reasoning
• Link: cs336.stanford.edu
• Assumes coding proficiency

Course 4 — CS236: Deep Generative Models

• Provider: Stanford
• Focus: Diffusion models, VAEs, flows, image synthesis
• Format: YouTube playlist

Milestone: Complete one generative AI project (image or text) with documented experiments

Block 3: Vision & Agents (Q3 — Jul to Sep)

Course 5 — Deep Learning for Computer Vision

• Provider: University of Michigan
• Focus: CNNs to modern architectures, detection, segmentation, generation
• Format: YouTube playlist

Course 6 — LLM Agents

• Provider: UC Berkeley
• Focus: Planning engines, tool use, reasoning systems
• Format: YouTube playlist

Milestone: Build a simple LLM agent with tool use

Block 4: Reinforcement Learning & Systems (Q4 — Oct to Dec)

Course 7 — Reinforcement Learning

• Provider: DeepMind × University College London
• Focus: Policy optimization, value learning, planning
• Format: YouTube playlist

Course 8 — MLSys Seminars

• Provider: Stanford
• Focus: System architecture, productionization, performance optimization
• Format: YouTube playlist

Milestone: Deploy one ML model to production with proper monitoring

Resources

• Primary: Course playlists as listed above
• Supplementary book: Build a Large Language Model from Scratch (if not completed in 2026)
• Practice: Kaggle, personal projects after each block
• Study Cadence: 8–10 hours/week

Integration Points

• Connect AI systems knowledge to AWS infrastructure (2026 Stream 1)
• Apply privacy engineering lens to ML pipelines (differential privacy, federated learning)
• Use philosophy stream to reflect on implications of each technology studied

Study Stream 2: Philosophy & Humanities (continuation)

Carry forward the 2026 monthly book club format. Books TBD based on what was completed in 2026 and emerging interests.

Study Cadence: 6–8 hours/week

Study Stream 3: Privacy Engineering (continuation)

Deepen 2026 work on ZKPs, TEEs, and digital identity. Focus shifts to privacy in ML contexts.

• Federated learning
• Differential privacy in practice
• Privacy-preserving inference

Study Cadence: 3–4 hours/week

Quarterly Milestones

Q1 (Jan – Mar)

• Complete MIT 6.S191
• Complete Karpathy Zero to Hero
• Milestone: Implement a small LM from scratch

Q2 (Apr – Jun)

• Complete Stanford CS336
• Complete Stanford CS236
• Milestone: One generative AI project documented

Q3 (Jul – Sep)

• Complete UMich Computer Vision
• Complete Berkeley LLM Agents
• Milestone: LLM agent with tool use

Q4 (Oct – Dec)

• Complete DeepMind RL Course
• Complete Stanford MLSys
• Milestone: ML model deployed to production

Success Criteria

• ✓ 8 courses completed
• ✓ 4 hands-on projects (one per block)
• ✓ Philosophy reading continued year-round
• ✓ Privacy engineering deepened in ML context