Phase 1 — Python & Math
April 18, 2026
Get the learning environment set up and the roadmap locked in
What I Did
Researched and finalized the Python & AI/ML learning roadmap
Decided on PyTorch over TensorFlow as the primary deep learning framework
Selected courses to use: Mathematics for ML, ML Specialization, and Deep Learning Specialization (all deeplearning.ai) — dropped the IBM courses
Picked books: Python Crash Course, Hands-On ML (Géron), Deep Learning (Goodfellow et al.), Mathematics for ML (Deisenroth et al. — free PDF)
Set up Logseq as learning journal, connected to a private GitHub repo for version control
Configured the Git plugin for auto-commit and sync
What I Learned
PyTorch is the dominant framework in research and modern ML — TensorFlow has lost the research community
The Hands-On ML book is still worth using despite the TensorFlow title — Keras is now framework-agnostic and the sklearn chapters alone are worth it
Privacy-preserving ML (concrete-ml, PySyft) is a natural extension of existing ZK/crypto background — treat it as an expansion track after core ML foundations are solid
Logseq works on plain markdown files, so GitHub is a cleaner sync solution than Google Drive for a developer workflow
Bugs & Blockers
None today — setup day
Concepts That Need More Time
What a tensor actually is and why it's the core ML data structure — need to sit with this before Phase 1 starts
Tomorrow
Start Mathematics for ML course (deeplearning.ai)
Implement a dot product from scratch in plain Python — no numpy
Read Python Crash Course multiple chapters
Wins
Roadmap is locked, environment is set up, journal is live on GitHub
Didn't overthink the tooling — picked Logseq and moved on