Projects beat certificates for AI roles. Each below maps to a skill interviewers probe. Build 2–3, deploy them, write a clear README.
Beginner (classic ML)
- Spam/sentiment classifier — text → label with scikit-learn. Shows the full ML workflow. (start here)
- House price / marks predictor — regression + a simple web UI. Shows regression + deployment.
- Customer segmentation — clustering on real data. Shows unsupervised learning + EDA.
Intermediate (deep learning & LLMs)
- Image classifier — CNN on a custom dataset (your own photos). Shows PyTorch + CNNs.
- "Chat with your notes" RAG app — the flagship. Shows embeddings + RAG + product sense. (guide)
- Resume → structured JSON parser — LLM + validation. Shows practical LLM engineering.
- Semantic search over a blog/docs — embeddings + vector DB.
Advanced (stand out)
- AI research agent — tool-using agent that searches + summarises. Shows function calling + agentic loops.
- Fine-tuned small model for a niche task — shows you understand when/how to fine-tune.
- AI feature in a real app with cost control, streaming and evals — shows production maturity (the rarest, most valuable signal).
What turns a project into an offer: a live demo link, a README explaining the why and the trade-offs, and honesty about limitations. "Here's what it does, how it works, and what I'd improve" beats a flashy demo with no depth. Deploy free: hosting guide.