Today I'm walking you through turning a CSV export into an interactive dashboard in Claude Code. Charts, filters, summary cards, all in one file you can double-click, poke at, and email to your team.

You'll have Claude profile the data, build the dashboard from one prompt (it's below, ready to paste), then refine it in plain English. It only takes about ten minutes end to end.

Let's get into it 👇

See the whole platform. No guided tour.

Skip the sales call. Walk through Gladly's interface yourself — the AI suggestions, the unified customer view, the full conversation thread. 15 minutes, no installation, no commitment.

A spreadsheet answers questions one cell at a time. You scroll, you squint, you build a pivot table, and twenty minutes later you have one answer and nine new questions.

A dashboard answers at a glance. Revenue trend, top performers, the weird dip in March, all visible the second it loads, with filters to slice it any way you want.

Until recently, getting one meant business intelligence software (think Tableau or Power BI: powerful, expensive, and someone has to be trained to drive it). 

Now a dashboard is something you describe, and Claude Code builds it while you watch.

The output is a single HTML file (that's a web page saved as a regular file). Double-click it and it opens in any browser, works offline, and travels by email or Slack like any attachment. 

Whoever you send it to double-clicks and gets the full interactive dashboard.

What you need:

Three things, and you likely have all of them.

The Claude desktop app, opened to the Code tab. Same agentic engine as Cowork, pointed at files. And relax: the Code tab speaks plain English. You will type sentences, and it will write every line of code.

A CSV file. CSV is the plain-text spreadsheet format, and it's the universal export: your CRM, your bank, Shopify, Google Analytics, your email platform, and Excel itself all produce one. Any "Export" or "Download data" button almost certainly hands you a CSV.

A fresh folder. Make a new folder, drop the CSV in it, and point Claude Code at that folder when it asks where to work. Keeping it separate means Claude only sees the data you brought.

Step 1: Make it read before it builds

The mistake beginners make is asking for the dashboard immediately. Ask for a profile of the data first:

"Read sales_2026.csv and give me a quick profile: what each column contains, the date range, the row count, and anything messy, missing, or duplicated."

Thirty seconds later you'll know your export has 4,812 rows, that "Region" is blank for 6% of them, and that March appears twice because you exported it twice.

This step does two jobs:

  1. You catch the messy stuff before it becomes a misleading chart

  2. Claude now understands the data well enough to make smart choices about what deserves a chart at all.

Step 2: One prompt, one dashboard

Now the build. 

Tell Claude what you want in plain language:

  • Summary cards for the headline numbers

  • Charts for the trends

  • Filters for the ways you naturally slice the data

  • Everything in one HTML file. 

The Prompt Drop below is the full copy-paste version.

Then watch it work. Claude will write the page, embed your data in it, and tell you when it's done. A file called dashboard.html appears in your folder.

Double-click it.

You'll see big number cards across the top, a trend line, and bar charts comparing categories, with dropdown filters that update every chart as you click. 

At the bottom sits a searchable table of the raw rows for when you need to chase a specific record.

Step 3: Edit by describing

The first version will be 80% right, and closing the gap is the fun part. Try giving it some instructions for iteration:

"Make the revenue chart monthly instead of weekly."

"Add a filter for sales rep."

"Swap the pie chart for a bar chart, and put the biggest category on top."

Three or four rounds of this and the dashboard fits your data like it was built by hand (which, technically, it was).

Bonus: make it self-updating

One extra instruction turns this from a one-off into a system: 

"Set this up so the dashboard pulls in every CSV in this folder. When I drop in a new file and say 'refresh,' rebuild it with the new data, same design."

Now next month's export is a drag, a drop, and one word. 

One caveat before you hit send

Your data gets embedded inside the HTML file. 

That's what makes it portable, and it also means anyone who has the file has every row in it, including whatever you filtered out of view. 

That’s fine for a team dashboard, but think twice before attaching customer data or salary data to an email (exactly as you would with the raw spreadsheet).

Paste this into Claude once your Gmail is connected, fill in the blanks, and save it in the project or conversation where your triage runs. 👇

I have a CSV file called [filename.csv] in this folder.

First, read the file and give me a quick profile: what each column contains, the date range, the row count, and anything messy, missing, or duplicated. Flag problems before building anything.

Then build me an interactive dashboard as a single HTML file called dashboard.html that works offline in any browser. Include:

1. Summary cards across the top for the headline numbers

2. Charts showing the main trends and comparisons in this data

3. Filters so I can slice everything by [date, category, region...]

4. A searchable, sortable table of the raw rows at the bottom

Embed the data directly in the file so it travels as one attachment.

Keep the design clean and readable.

When you're done, tell me the three most interesting things you noticed in the data.

Run it manually for a few days, tune the rules, then schedule it. The agent gets exactly as good as the rules you give it.

Think You Know What AI Does Next?

Which model leads the next benchmark? Which AI lab ships the next major breakthrough?

Kalshi lets you trade on real-world AI and technology events as the industry moves. If you follow launches, model updates, and benchmarks closely, put that knowledge to work.

Bonus credit varies from $15 to $500. Terms apply.

Your roundup of the latest model releases and updates from the biggest AI labs.

  1. Google introduces Gemini Omni. Google's new model family creates and edits video from any input, and you edit with natural language: describe the change and it happens, the same pattern as today's dashboard workflow. The first model, Gemini Omni Flash, is rolling out to Google AI Plus, Pro, and Ultra subscribers, and free inside YouTube Shorts and the YouTube Create app. (Google)

  2. Meta ships Muse Image. The first image model out of Meta Superintelligence Labs, built on its Muse Spark flagship, is live in the Meta AI app, Instagram Stories in the US, and WhatsApp in some countries. It ranks No. 2 on Arena (the human-preference leaderboard) for image generation and editing. The friction point: it draws on Instagram for social context, and users are already pushing back over their photos feeding the model. (TechCrunch)

  3. GPT-5.6 Sol posts benchmark wins over Claude Fable 5. OpenAI's upcoming flagship leads on Terminal-Bench, a test of exactly the kind of agentic computer work in today's edition, scoring up to 91.9% against Fable 5's 84.3%, at roughly half Fable's price. Worth the grain of salt: on SWE-Bench Pro, the benchmark many engineers treat as closer to real work, Fable 5 still holds the top published score, and Sol remains in a government-vetted preview you can't use yet. (TechTimes)

Advertise with Build with AI

Get in front of an audience of professionals using AI day-to-day: founders, engineers, operators, and product builders.

Interested in advertising? Respond to this email for rates and details.

Until next time,

William Ryan

Editor-in-Chief @ Build with AI

PS: Follow me on X for daily updates and AI workflows.