03 ML Fundamentals & Applied ML

What to do Complete all lectures and homework of fast.ai's Practical Deep Learning for Coders Part 1. Commit every homework notebook to a dedicated public repository. The course uses PyTorch and takes a top-down "code first" approach — train models before learning the maths behind them.

What counts as complete All lecture homework notebooks committed. One substantive personal project using techniques from the course (image classifier, text classifier, or tabular model) shipped.

What to do Either enter an active Kaggle competition and submit a scored entry, or publish a substantive Kaggle notebook analysing a public dataset. Both count. The notebook route is usually better value for learning since it rewards explanation, not just leaderboard rank.

What counts as complete Public Kaggle profile with at least one entry or notebook linked from this portfolio.

What to do Either deploy a Gradio or Streamlit demo to Hugging Face Spaces (free tier), or fine-tune a small open model (GPT-2 small, Phi-3 mini, or similar) with LoRA on a task-specific dataset and upload the adapter to the Hub.

What counts as complete Public Hugging Face profile with at least one Space or model linked from the portfolio.

What to do Audit the three-course Machine Learning Specialization on Coursera (free, no certificate). Covers supervised learning, advanced algorithms, and unsupervised/recommender/reinforcement learning. Theory-first complement to fast.ai's practical-first approach.

What counts as complete All three courses' lectures watched, all programming assignments attempted and committed to a public repository.

What to do Build and deploy one end-to-end ML project: data pipeline, model training, evaluation, deployment as a live demo. Publish a substantial writeup covering problem, dataset, approach, evaluation methodology, and limitations.

What counts as complete Live demo reachable at a public URL, training code and notebooks in a public repository, technical writeup published on the portfolio.

What to do Follow Andrej Karpathy's Zero to Hero lecture series and build along — micrograd (tiny autograd engine), makemore (character-level language model), and ideally the GPT-from-scratch lecture. This is the "understand the fundamentals" course — the autograd, backprop, attention maths all built from Python and numpy.

What counts as complete Working micrograd or mini-GPT implementation in a public repository with commentary annotations showing understanding of each component.

What to do By Phase 3, the senior-engineer advocate contact should be actively making introductions — to other engineers, hiring managers, or interesting open-source projects. If not, diagnose why and adjust.

What to do Optional but recommended: apply to AI data-labelling or tutoring platforms (Outlier, Scale, Surge, Mercor) as a part-time income source. The work exposes you to production model behaviour and adds modest income. Skip if the application friction outweighs the value.

What to do Add approximately fifteen connections from the ML and AI community — Kaggle peers encountered during ML-02, fast.ai cohort members, Hugging Face contributors whose work has been referenced, and practitioners publishing applied ML writeups. Anchor connection requests to ML-02, ML-03, or ML-05 artefacts. Use template 4 in content/linkedin-templates.md ("senior engineer whose content has been engaged with") where applicable; otherwise template 3.

What not to worry about

• Cold academics — this cluster is for practitioners, not researchers.
• Low acceptance rates from very senior or high-profile accounts; the connection attempt itself is the signal to the algorithm.

What counts as complete Fifteen personalised connection requests sent, anchored to specific ML artefacts shipped in this phase. Outcomes tracked privately.

Why this phase ML artefacts (Kaggle notebook, Hugging Face Space, applied ML project) provide shared-domain context that engineers-only references from earlier phases cannot. This is the phase where the programme first reads as an AI engineer's profile, not a web engineer's.

What to do Professional LinkedIn post at Phase 3 close. Template in content/linkedin-templates.md. Lead with the applied ML project and specific techniques used.

What to do One LinkedIn post per shipped ML artefact (Kaggle, Hugging Face, applied ML capstone). Each focuses on problem, approach, and one non-obvious insight.