8‑Week Execution Plan (detailed)

Weeks 1–2 — Foundations

• Notebook 1 (classification): train/val/test; acc; confusion matrix
• Notebook 2 (regression): MAE/MSE/R²; learning curve
• Repo hygiene: tests for helpers; CI green
• Post: “How I’ll learn AI as a builder”

Weeks 3–4 — LLM/RAG & AI Search (DocChat)

• Ingest→chunk→embed (BGE‑M3)→vector index (FAISS/Qdrant)→rerank (Rerank‑3/Nimble)→answer with citations
• Log p50/p95 & token counts; RAG Triad eval (context relevance, groundedness, answer relevance)
• Deliver: /ask endpoint + minimal UI; 60‑sec demo; README diagram

Weeks 5–6 — Recommenders (Mini‑Recsys)

• Candidate gen (FAISS ANN) + simple ranker
• Eval: precision/recall/nDCG@k; beat popularity by ≥ 0.05 at nDCG@10
• Deliver: /recommend endpoint; metric table + plot

Week 7 — Vision/AR/Robotics (one small win)

• Near‑duplicates or quality scoring (CLI+API; 30‑sample QA), or a Unity AR Foundation demo

Week 8 — Decide & Commit

• Score lanes (enjoyment, market pull, speed, depth, defensibility)
• 1‑page memo + 90‑day roadmap; public announcement