The fastest way to actually learn Generative AI is to build something you'd want to use yourself. Here are 12 project ideas across three difficulty tiers, each chosen to practice a specific skill — not just "make a chatbot" for the fourth time.
Beginner projects
Good if you've just learned to call an LLM API and want to practice prompting before adding any real complexity.
Email / message summarizer
Summarize long emails or Slack threads into 3 bullet points. Practices basic prompting and API calls — no framework needed.
Resume-to-bullet-points rewriter
Takes a rough job description and rewrites it into strong resume bullet points. Good practice for structured prompting and few-shot examples.
Study-notes Q&A bot
Paste in your own notes, ask questions about them. A stripped-down version of our AI Study Assistant project.
Recipe generator from ingredients
Input a list of ingredients, get 3 recipe suggestions. Fun, low-stakes way to practice prompt design and output formatting.
Intermediate projects
These require a real backend and usually one more moving piece — a vector database, structured output parsing, or basic evaluation.
Chat with your PDF
A full RAG pipeline over your own documents, with source citations shown in the UI.
Follow the RAG Guide →Customer support ticket classifier + responder
Classify incoming tickets by category and urgency, then draft a suggested response for a human to review — practices structured outputs and evaluation.
Code documentation generator
Point it at a codebase, generate docstrings and a README. Combines prompting with basic file parsing.
Meeting-notes-to-action-items tool
Transcribe (or paste) meeting notes, extract action items with owners and due dates as structured JSON.
Advanced projects
These combine multiple skills — agents, RAG, evaluation — the way a real production AI feature usually does.
Multi-tool research agent
An agent that can search the web, read a PDF, and do basic math to answer a research question — practices tool calling and the agent loop.
Read: AI Agents Explained →RAG system with evaluation dashboard
Build a RAG pipeline and a small dashboard tracking retrieval recall and answer faithfulness over a test set — this is what separates a demo from a defensible system.
Read the Evaluation Guide →AI career assistant
Analyze a resume against a job description, detect skill gaps, and generate a learning plan — the exact capstone project in our course.
See the capstone project →Voice-driven AI assistant
Chain speech-to-text → LLM → text-to-speech into a voice assistant for a specific task (scheduling, Q&A, note-taking).
Every project above maps to something taught hands-on, with instructor feedback, in our Generative AI course — if you'd rather build these with structure and feedback than entirely alone, that's exactly what the course is for.
How to pick one
- Pick something you'd actually use. Personal stakes (your own notes, your own resume, your own workflow) drive you to finish it — this is deliberate in how our own course projects are designed.
- Constrain the scope on day one. "An AI agent that can do anything" never ships. "An agent that can look up one API and do basic math" does.
- Ship an ugly version fast, then improve it. A working end-to-end pipeline with a plain interface beats a polished UI wrapped around nothing that actually works yet.
- Write it up properly when you're done. Problem → architecture → tech used → outcome. That format is what makes a project readable on a resume or in an interview — see our interview questions guide for how these projects get discussed in practice.
Keep learning: See real student projects in the projects gallery, or build yours with feedback in the Generative AI course.