AI Hints
Home›Generative AI Tutorial›Ten Mini Projects
Generative AI · Lesson 77 of 77

Ten Mini Projects

Start building this week.

⏱ 2 min read▶ Video at 2:45:28🎓 Free
Watch this lesson in the Generative AI courseWatch on YouTube ↗Full Generative AI course

The best way to learn is to build. In this lesson, you will get ten mini projects, from very easy, to more ambitious. For each one, you will see the idea, the tools it uses, and a simple way to start. Pick one, and begin this week.

Ten Mini Projects
From the video: Ten Mini Projects
  • Ten projects, easy to ambitious
  • The idea and the tools
  • A simple way to start

Projects one to three: no code needed

The first three need no code. Project one, a personal study coach. Paste your notes, and ask the assistant to quiz you, and to explain your mistakes. Project two, a meeting summariser. Paste a transcript, and ask for decisions, owners, and deadlines. Project three, a writing studio. Draft your emails, and posts, with a prompt template that captures your tone of voice.

1. Study coachpaste notes, get quizzed
2. Meeting summariserdecisions, owners, deadlines
3. Writing studioa prompt template for your voice

Projects four to six: a little code

The next three use a little Python. Project four, an email sorter. Send each message to a model, and have it return J S O N, with a category, and a priority. Project five, a document chat. Split your files into chunks, embed them, and answer questions, using retrieval. Project six, a daily briefing bot, that gathers a few web pages, and summarises them, every morning.

4. Email sorterJSON with category and priority
5. Document chatchunks, embeddings, retrieval
6. Daily briefing botsummarise a few pages each morning

Projects seven to ten: more ambitious

The last four stretch you. Project seven, a research agent, that searches, reads, and writes a short report, with sources. Project eight, a support assistant, for a small shop, grounded in its real policies, with a hand over to a human. Project nine, an image workflow, that creates matching pictures, for a whole product line. Project ten, a local, private assistant, running a quantised model, on your own laptop.

7. Research agentsearch, read, write with sources
8. Support assistantgrounded in real policies
9. Image workflowa consistent look for a product line
10. Private local assistanta quantised model on your laptop

How to run any project

Follow the same method for every project. Define the job in one sentence. Collect ten real examples. Build the smallest version that works. Test it on the examples, and score the results. Fix the biggest problem first. Add guardrails, privacy, and human review where needed. And share it, because feedback makes it better. Small, tested, and finished, beats big, and unfinished.

  1. One-sentence goalwhat job does it do?
  2. Ten real examplesyour test set
  3. Smallest versionget it working
  4. Test and scorefix the biggest problem
  5. Guardrails and reviewprivacy and safety
  6. Share itfeedback makes it better

Summary

Wonderful. Ten projects, from a study coach, to a private local assistant, and one method to build them all. Choose one today, finish it, and then pick the next. You have the knowledge, so now enjoy building.

Key takeaways
  • Start with no-code projects
  • Then a little code, then agents
  • Small, tested and finished wins
Next up: Quiz Time — test yourself