MP3 showcase and final project launch

Tuesday, 11/17 · Week 12

Final Project Proposal Due Sunday, 11/22

Post-Class Notes

TL;DR#

Today was for seeing Mini Project 3 side by side with projects that do something similar, finding where each one breaks, and combining the APIs in the class with AI as a first step toward your final project.

What to do next#
  1. The final project: we will write three ideas under ## Final project ideas in logs/wk12.md, one sentence each, and let your agent question you about them with the prompt below, in class on Thursday 11/19. The Proposal is due Sunday 11/22 at 11:59 PM.
  2. Mini Project 4 is due Sunday 11/29 at 11:59 PM, and you will demo it on Tuesday 12/01.
  3. Checkpoint 6 is due Sunday 11/29. It now includes logs/wk12.md in your oim3690 repo, with your sentence from today under ## API mashup. Use the heading exactly as written.
  4. The OpenRouter class key: the old key stops working on Friday 11/27. If a project of yours uses it, replace OPENROUTER_API_KEY in Vercel's settings with the key on the OpenRouter Class Key page and deploy again.
The AI feature in your final project#

Every final project must have an AI feature: at least one feature calls a model with instructions you wrote for this application and the person it is for, so that its answers fit your product. The spec asks you to name that person in your README.md. A chat box counts when its instructions do that. A chat box that passes what a visitor types to a model with no instructions does not.

The project goes in a new public repo, named in lowercase with hyphens after what it does, such as study-room-finder. Full reference: Final Project, under Requirements.

Your final project ideas#

Your three ideas can come from:

  • Your API mashup from today
  • Your Mini Project 3, grown with an AI idea your group gave you. Sign in on the vote page to read the feedback written for you under Feedback on your project
  • Your own life, a club, a job, or your Mini Project 1 client's Later list

If you already know what you want to build, write it down with two other ideas to compare it against. An idea grown out of Mini Project 1, 2 or 3 must be significantly different from it: every feature in PROPOSAL.md must be one that mini project does not have.

Then paste this into your agent, with your ideas in place of the brackets. If you will miss class on Thursday, do it on your own before the proposal is due:

Here are my three ideas for the final project: <your three ideas, one sentence each>. Ask me one question at a time about who will use each one, what they need, and where its data comes from. Do not pick one for me and do not suggest new ideas.
Combining APIs with AI#

The sentence from today's mixed groups: A tool for ___ that takes ___ from ___ and ___ from ___, and asks AI to ___.

For example: A tool for my roommate that takes tonight's forecast from Open-Meteo and recipes from TheMealDB, and asks AI to pick one dish that suits the weather. APIs in the Room lists every API the class used, with a link to its docs and whether it needs a key. A key never goes in your repo.

Breaking your own site#

The cases from today work on any project that calls an API, including your final project:

  • The API fails. Load your site first, then in DevTools open Network and change No throttling to Offline, and do the main action. Switching to Offline before the page loads shows only the browser's own offline page. Set it back to No throttling when you are done.
  • An input that does not exist, such as the city Atlantis.
  • An input outside the data, such as a date 20 days ahead on a forecast that covers 16. If your date picker will not let you choose it, find the line in your code that sets the limit and check that it matches what the API returns.
  • A search with no results, such as zzzz.
  • An input much larger than you tested. One road trip planner worked from Boston to Mexico City, then could not find stops along a route that long.

Predict what your site will do before you try each one. If it surprises you, ask your agent why, and fix it. The same cases make good tests for the states in your final project's PROPOSAL.md.

Avoiding the generic AI look#

The lightning discussion asked what makes a page look AI-generated. The usual tells are an indigo-to-violet gradient, fonts nobody chose, and cards and rounded corners on every section. The Generic AI Look has the checklist from class and the full catalog. To keep your final project from looking that way:

  • Write your design rules down before your agent builds, and point it at them: Write the System Down
  • Pick a few sites you admire and say what you take from each: Learn from Sites You Admire
  • Use your site yourself before anyone else does. You are its first user.

The other question from that discussion, whether a site that runs is safe, is one your final project answers in part: no key in the repo, try/catch around each fetch, and Row Level Security on every Supabase table.