Data & AI Club · Workshop 1

Follow along.

Everything we do today, in plain words, in order. Keep this open on the side and work down it. You can do all of it again at home from this page alone.

The AI sounds confident whether it's right or wrong. Reality is the judge. Your job is to check.

0  Before you start

Ten minutes. Get these in place and everything after is smooth.

  1. Open the Claude desktop app (from claude.com/download) and sign in. On a paid plan you'll see three tabs: Chat, Cowork, Code. No paid plan? Use claude.ai in your browser and follow the same steps in a Project.
  2. Make a folder called Career Fit — your Desktop or Documents both work. Put three things in it: your CV as a PDF (Optional: North Star, Performance Reviews, Transcripts, etc...), the Career-Vault-Builder.md, and the Internship-Radar-Prompt.md files from the packet. Everything today lives in this folder. On Windows, if a step ever says it can't see the folder — which can happen when your Desktop is synced to OneDrive — click Add folder in Cowork and pick it. That fixes it.
  3. Be signed into LinkedIn in your browser. Task 2 reads your job alerts.
Stuck on any step, ever? Paste the step and what happened into any chat and ask it how to do that exact thing. That's not cheating — that's the skill.
Before Thursday, clear two sign-ins that can quietly stall a run. Each sends a notification that's easy to miss, and the run just waits until you handle it. Email: connecting Gmail or Outlook (Cowork → Customize → Connectors) is a one-time approval and the smooth path — the radar then reads your alerts with no browser login at all. Signing into Gmail in Claude's built-in browser instead can trigger a "confirm it's you" from Google. LinkedIn: only matters if you turn on the optional "open each posting" step later — the first time, LinkedIn usually sends a "verify this device" code to your email or the LinkedIn app. Do both now, with your phone within reach.

1  The one habit: direct → execute → verify

Everything you do with AI is the same three steps, whether you're asking a question or building an app:

A model is a very good guesser of the next words. It has no part that checks its guess against the world, so a right answer and a wrong one sound exactly the same. And it only knows what you put in front of it — your message, the files you gave it, a page it can open. Not your inbox, not the internet, unless you hand it over. That's why you check.

2  Task 1 — Find out what fits you

About 20 minutes. You'll end with a one-page profile the rest of the day uses.

Task 1, step 1 — in Cowork (the tab, not Chat), attach your Career Fit folder. It shows as a chip under the box; that’s the only folder it can see.
  1. In Cowork, point it at your Career Fit folder. (It can only see this folder — nothing else on your computer.)
  2. Paste this and send it:
    Paste into Cowork

    Read the Career Vault Builder file and my CV. Give me the questionnaire, short version.

  3. Answer its questions, one line each. "I don't know" or "skip" is a fine answer. Don't overthink it.
  4. Read what it gives back: it reflects your strengths, names your gaps, and proposes two or three roles you'd fit — including some you didn't think of. Push back, refine, tell it what lands and what doesn't. This is the useful part.
  5. Check one thing. Pick one fact it stated about a company or a role — a location, a pay range, how you'd get in — and look it up. Did it hold? Tell it if it didn't.
  6. Save it. Paste:
    Paste into Cowork

    Write my Career Fit profile as one page. Save it as career-fit.md in this folder.

    Watch it ask before it writes. Say yes.
Done looks like: a file called career-fit.md sitting in your Career Fit folder, with your target roles, sectors, and locations listed in it. That file is what Task 2 reads.

3  Task 2 — Build the thing that works while you don't

About 20 minutes. An internship radar: it watches for new jobs and ranks them against your profile, every week (or every day), on its own.

First, set a LinkedIn job alert

This is the feed your radar reads. Six steps, about two minutes — the titles and locations come straight from the Career Fit profile you just built.

  1. 1. Sign in to LinkedIn, then click Jobs in the top bar.
    LinkedIn step 1
  2. 2. Click Add job preferences.
    LinkedIn step 2
  3. 3. Add the job titles and locations your Career Fit profile suggested, and set Employment type → Internship. Save & continue.
    LinkedIn step 3
  4. 4. Choose who can view your preferences. Recruiters only is fine. Save & continue.
    LinkedIn step 4
  5. 5. Tick the alerts you want (one per title and location). Save.
    LinkedIn step 5
  6. 6. Leave Email updates: On, then Update settings. This is the point — the alerts arrive as emails your radar reads.
    LinkedIn step 6

The first email can take up to an hour, so it's fine if it hasn't arrived by the end of class. You're building the automation; it runs whenever the emails come.

Let Claude reach your inbox. The radar reads the alert from your email, so give it a way in: connect your email in Cowork → Customize → Connectors (Gmail or Outlook), or just be signed into your webmail in Claude's browser when it asks. Either works.

Then build the radar

Everything for Task 2 is in the left sidebar. Scheduled is where you build the radar; Customize is where you connect your email.
Task 2, step 1 — on the Scheduled page, click New task → Create with Claude.
  1. In Cowork (same folder) go to Scheduled → Create with Claude.
  2. Paste this:
    Paste into Create with Claude

    Every Monday morning, search my email for the newest job-alert email from jobalerts-noreply@linkedin.com. That email lists new job postings — each with a role, employer, location and a short description right there in the email. Read career-fit.md in my folder. Rank the postings in the email best fit first, dropping any that break a hard constraint from that profile. Give me a table: role, employer, location, why it fits or doesn't in two lines, and the link. When you email me the summary, write it as a numbered list with each posting on its own line (rank. Role — Employer, Location: the two lines), never one paragraph, because email flattens a table into a wall of text; list any dropped postings under it. Also save the full result on disk so I keep a history: in my Career Fit folder, make a subfolder Internship Radar/[alert date] — [number of postings] matches (for example Internship Radar/2026-09-16 — 6 matches) and write the full ranked table, with links and the dropped postings, as ranking.md inside it. Never move or change career-fit.md — it stays at the folder root. Rank from what the email itself says — you don't need to open the postings. Optional, only if a posting looks like a strong fit: open its link for the full description and refine that one row. If a page asks you to sign in or won't open, skip the deep read and rank that posting from the email instead. Never invent a detail you didn't read. If you can't find the alert email at all, tell me — don't guess.

  3. Make sure your email is reachable — connect Gmail or Outlook in Cowork → Customize → Connectors, or be signed into your webmail in Cowork's browser (the tip above). The radar reads the alert straight from your inbox, so it needs no LinkedIn login to build the table.
  4. Run it now. It should read your newest alert email and hand you a ranked table off the postings listed in it.
Reading the run right. If a run shows "succeeded" in a few seconds, it did nothing — it just launched. Open the run and read what it actually did. A real run takes a few minutes. It only opens each posting's page when you asked for the optional deep read, and if a posting page throws up a sign-in wall it simply ranks that one from the email instead of stopping. If no alert has landed in your inbox yet, the run has nothing to rank — that's fine, it'll work once your first alert arrives (which can take up to an hour).
Three small things. The scheduled radar runs on your laptop, so keep it on with the Claude app open at the run time. If a run looks wrong, edit the existing task — don't create a second one, or you'll get duplicate emails. And the schedule runs on UTC: a 12:00 Barcelona task runs at 12:00 now, but at 11:00 from late October to late March (you can nudge it back then). Free plan? Scheduled tasks need the paid desktop app — pair up with someone who has it for this task, and you can still build the Code-tab version below on your own.
Not into internship hunting? Two other doors, same folder: build a skill (Cowork → Customize → Skills) that does a weekly task in your style, or open the Code tab and ask it to "build the internship radar as a page I run with one click" — the point there is watching it plan and show you the changes before it makes them.
Done looks like — a saved task on your Scheduled page, like this:
A finished scheduled task: Daily LinkedIn job-alert ranking
Done looks like: a saved task on your Scheduled page that runs on its own, and one table it produced now.

4  Task 3 — Prove it works

About 15 minutes. The whole workshop in one act.

  1. Look at the table. Do the top three even look like jobs you'd want? If not, your profile is off — fix career-fit.md, not the tool.
  2. Open the top posting yourself. Does the real posting say what the two lines said? Deadline, location, the actual role.
  3. Ask it why. "What put this one first?" If the answer is vague, it guessed.
  4. Change one line of your profile — a city, a sector — and run again on the top three. Did the ranking move the way it should?
What would you have believed if you hadn't looked?

5  Words you'll hear

Prompt What you type. A good one is a brief: who it should be, what it needs to know, the task, what done looks like.
Hallucination A confident guess said as fact. It doesn't feel any different from the inside — so you check.
Cowork The tab where you hand Claude a folder and it does a job, asking before it acts.
A skill A saved way of doing something — your tone, a format, a checklist — so you don't re-explain it.
Scheduled task A job that runs on its own on a schedule, in the cloud or on your laptop when it needs your logins.
Project A chat with a memory: files you upload once and it keeps reading. Works on the free plan.
.md A plain-text file with light formatting. Your career-fit.md is one. Opens in Notepad.
Career-fit.md The one-page profile from Task 1. Everything else today reads it.

6  Keep going

Take five minutes and list, from your own week: one task you could automate (it runs without you), one to augment (you and it together), one to keep human (yours on purpose). The first one is your next build.

Continue at Home: your Career Vault

Your one-page career-fit.md is the seed of something bigger, and the whole deeper build is optional and only at home — nothing in class touches it. When you're ready, open the Career-Vault-Builder.md file again in Cowork on your folder (or paste it into any AI) and say "build the full vault." It runs a longer questionnaire, has a real back-and-forth with you, then turns your profile into a linked Obsidian vault: a note per skill, experience, story, target career and hard constraint, all cross-linked so opening any one shows the whole web — which skills a role wants, which stories to lead with, which constraints rule it in or out.

To see the connections, install Obsidian (free, obsidian.md) and open your Career Vault folder as a vault — the graph and backlinks light up. If you'd rather not, the plain files still read fine in any text editor.

And the one line to carry out the door: when it's confident, check it. Not because the tool is bad — because it's fast and literal, and it will never tell you which of its sentences to doubt.