Vibe Coding for Language Data Visualization

Charles Lam, Language Centre

2026-02-25

Why Interactive Dashboards?

How we usually visualize data

  • Static charts in papers or slides
  • Screenshots from Excel or SPSS; Output from
  • Fixed perspectives (the reader sees only what we chose to show)
  • Linear storytelling (one path through the data)

Benefits for Research Communication

  1. Readers can explore on their own, enhancing transparency
  2. This can better engage readers with hands-on interaction
  3. Reusable: Keep the same tool for different (sub)sets of data (we typically have the same set of methods and visualization across projects)

Benefits for Teaching & Learning

  1. Immediate response with users adjust parameters
  2. Scaffold discovery without over-prescribing
  3. Data-Driven Learning (DDL) — Students explore corpus data directly

The “Vibe Coding” Boom

What changed in 2025?

  • AI assistants can generate larger chunks of working code from natural language
  • No-code/low-code platforms reduce barriers
  • Free deployment options make sharing easy
  • The barrier is no longer technical — it’s our own imagination

Vibe Coding = Coding by Vibes

The term (coined by Andrej Karpathy, Feb 2025)

“There’s a new kind of coding I call”vibe coding”, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.”

Source: [https://x.com/karpathy/status/1886192184808149383?lang=en]

. . .

In practice:

  • Describe what you want in plain English (or others languages really)

  • AI generates the code

  • You refine through conversation iteratively**

  • Deploy without deep technical knowledge???

  • “Forget that the code even exists” is not impossible, but knowing the logic / structure / code is actually quite desirable for speed and precision.

  • AI-assisted coding does help learning how to code

Why It Matters for Language Researchers

  • Textual data are not always transparent for data scientists
  • Think narratives – the story you want to tell in the research

Demo of Live Applications

Demo 1: Biomedical Science Corpus Dashboard

Qualitative Methods Dashboard

A data dashboard for exploring a corpus of academic writing in biomedical science.

Built with Replit (plus some Python)

. . .

Visit:

qualitiative-methods-dashboard.replit.app

Click “Use Biomedical Dataset” to start exploring.

Demo 1: Key Features

  • Browse corpus texts by discipline or text type
  • Search for patterns across the collection
  • Visualize distributions of linguistic features
  • Compare subsets of the data
  • Export findings for further analysis

Demo 2: Move-Step Analysis Tool

A tool to display rhetorical progression in the Methods section of science writing.

Built with Google AI Studio

AnnotateLab

annotatelab-860580144856.us-west1.run.app

Click “Display annotation” (top right) — expect 4-5 seconds loading time.

Demo 2: Key Features

  • Move-Step Analysis — rhetorical structure visualization
  • AI-assisted annotation — automatic identification of moves
  • Visual progression — see how methods sections unfold
  • Pattern discovery — compare across texts

What These Demos Illustrate

Technical:

  • Web-based (no installation)
  • Mobile-friendly
  • Shareable via URL
  • Free to host

Pedagogical:

  • Exploratory learning
  • Real research data
  • Immediate feedback
  • Low-stake experimentation

Break (5 minutes)

Hands-On Practice

Getting Started with Google AI Studio

Why Google AI Studio?

  • Free to use (with Google account)
  • No installation required
  • Generates working prototypes

. . .

Access now

Go to aistudio.google.com and sign in with your Google account.

Exercise 1: Simple One-Liner Prompts

Try these examples (5 mins)

Copy and paste these prompts to see what AI can generate:

. . .

Non-language data:

“Create a simple web app that converts temperatures between Celsius and Fahrenheit”

“Build a to-do list app with the ability to mark tasks as complete”

“Build a visualizer to display quadratic equations.”

From exercise 1: Observe these aspects:

  • Speed — How fast can you get a working prototype?
  • Iteration — Try saying “make the chart blue” or “add a download button” after the first iteration
  • Limitations — What doesn’t work as expected?
  • Refinement — How specific does your prompt need to be?

Exercise 2: Create your own prompt

Come up with a tool that would tell your story, e.g.:

  • Keyword comparison tool
  • Collocation visualizer
  • POS tag distribution chart
  • Sentiment timeline

Use the many tricks in prompt engineering: Chain-of-Thoughts, metaprompting, references (with images or links)

Key Points for Effective Prompts

  1. What it does — Clear description of functionality and goal
  2. What data it uses — Input format (text? CSV? user upload?)
  3. What output you want — Charts? Tables? Images? Downloadable files?
  4. User interaction — Buttons? Filters? Text input?
  5. Visual style — Clean? Colorful? Minimalist?
  6. Reference — Include some links or images for inspiration

Example: A Well-Crafted Prompt

Vague:

“Make a word frequency tool”

Better:

“Create a web app where users can paste text into a box. Use a clean, professional design with a blue color scheme. When they click ‘Analyze’, show:

  1. A bar chart of the 15 most frequent words (excluding stopwords)

  2. A table with word, frequency, and percentage

  3. A download button to save results as CSV. ”

On finding References & Inspiration

Where to look:

  • Existing tools
  • Research papers — What visualizations have you seen in your field? You can also instruct the coding agent to use specific packages
  • Colleagues — Describe what you want to someone else
  • AI itself — Ask: “What are common ways to visualize corpus data? Give me 5 options for X.”

Troubleshooting Tips

When things don’t work:

  • Be more specific — Add details about what went wrong
  • Break it down — “First, just make the file upload work”
  • Ask for explanation — “Explain why the chart is not showing” (CoT prompting)
  • Try alternatives — “Use a different charting library”
  • Iterate — Refinement is part of the process

Sharing Your Work

What’s next (in our subsequent sessions)?

For quick demos:

  • Screenshot the result
  • Share the prompt as inspiration
  • Record a short video

For real deployment:

  • Export code to Replit
  • Use Streamlit Cloud
  • (Google Cloud Run)
  • GitHub Pages

Resources

Continue your journey

Thank You!

Share your experiments, questions, and ideas!

Email: c.lam@leeds.ac.uk

Next:

  • Up to 8 spots for 4-month subscription of Replit
  • Monthly(?) meetups to share progress and further develop the projects
  • Potential for publication