1 of 15

2 of 15

Disclaimer

Nothing in this workshop is the official position of, or endorsed by, Anthropic or SCSP.

3 of 15

What we'll do today

1

Watch me build a live website from a dataset in 15 minutes using Claude Code

2

Then you build your own and publish it to a public URL

3

We’ll share and discuss

SCSP AI+ Expo · May 9, 2026

4 of 15

What we'll do — in order

Roughly 75 minutes; we'll move through these in order

1

Welcome & context

2

Chatbot vs. agent

3

The dataset & data safety

4

Live build & deploy

5

Hands-on (your turn)

6

Showcase

7

Wrap-up & resources

5 of 15

Agents vs chatbots

An agent is a model running tools in a loop.

The model

picks the next action — which tool to use, what to read, what to write.

The tools

are everything it can use on your behalf: read files, run code, browse the web, send messages.

The loop

is try → observe → retry, until the goal is met.

It is not just bigger models — it’s better tools and tighter loops.

6 of 15

New paradigm new risks

Three things to know before you point an AI at your data

01

Opt out of training data

Claude.ai → Settings → Privacy → "Help improve Claude" toggle off.

02

Avoid the lethal trifecta

Untrusted input + private data + external action. If a prompt has all three, a prompt hidden in the input can trigger unwanted actions.

03

Never paste secrets

API keys, PHI, CUI. If a file contains them, ask Claude to ignore those lines. Use Claude Code to help generate synthetic data if possible.

7 of 15

How you direct an agent

Treat Claude like a forgetful intern. Specific instructions and context are required.

1

Explore

Look at the data. Gather context.

2

Plan

Propose what to build. Ask the user before writing.

3

Execute

Carry out the plan.

4

Iterate

Show the result. Take feedback. Refine.

Your first prompt today asks Claude to look first and then ask you questions to create a plan — that's this loop in action.

8 of 15

Today's dataset

Stanford SNAP Facebook ego networks · McAuley & Leskovec, NIPS 2012

4,039

anonymized people

88,234

friendships

10

ego networks

1

edge-list file

facebook_combined.txt.gz

0 1 0 2 0 3 1 4 1 5 ...

9 of 15

LIVE BUILD

Watch me

Dataset → working interactive site → public URL — in 15 minutes, in plain English.

If something breaks, here's the backup:

learning-journey-ai.github.io/scsp-workshop-demo

10 of 15

11 of 15

Don't have Claude? Grab a free month while supplies last

6

free Claude Pro 1-month codes

first come, first served

What to do:

1

Come up to the front to claim a code while we get started

2

If you miss out — pair with a neighbor running Claude

3

Either way, take the reference card home and try later

12 of 15

Pick your difficulty

All in plain English, all on your reference card

BEGINNER

  • Change the color theme or font
  • Add your name to the page header
  • Add a footer with a link to your GitHub

INTERMEDIATE

  • Add a degree-distribution chart
  • Make node size scale by degree
  • Add a search box for a person ID

ADVANCED

  • Detect & color communities (Louvain)
  • Compute & display clustering coefficient
  • Try a different SNAP dataset

13 of 15

Showcase — share your URL

1

Share links in https://docs.google.com/spreadsheets/d/1NFtKRzlB6n9YVeHiG9zCupY9W1Lmf5YvfwlwvKn2axk/edit?usp=sharing

2

A volunteer or two: walk us through what you built

3

What do you want to build next?

14 of 15

Keep building

What to read, watch, and try after today

code.claude.com/docs

Learn more about Claude Code

snap.stanford.edu/data

More public network datasets to point Claude at

resources.anthropic.com/hubfs/The-Complete-Guide-to-Building-Skill-for-Claude.pdf

Learn to build your own skills

dcai.guide

Find AI resources and community events in DC

15 of 15

Thank you.

Questions?

Nicholas Wagner nwagner@learningjourneyai.com Learning Journey AI