DEV Community

Cover image for I Built Cursor for Spreadsheets.. But What for?
Hassan Zahar Rifat
Hassan Zahar Rifat

Posted on

I Built Cursor for Spreadsheets.. But What for?

WLH Challenge: Building with Bolt Submission

๐Ÿš€ Why I Built It

I manage a side project with a customer base, and like a lot of solo builders, I frequently use Google Sheets to keep track of metrics, revenue, and day-to-day data.

Over time, I found myself doing the same repetitive tasks โ€” writing formulas, cleaning up tables, copying logic across rows and it started to feel inefficient. Not difficult, just unnecessarily manual.

Thatโ€™s when I realized I didnโ€™t want to build another product just for the sake of the hackathon. I wanted to build something that I would actually use, something would solve a real business problem.

So I scrapped my original idea and started working on a spreadsheet that behaves more like an assistant. One where I could type plain language and get back working formulas, insights, or even full summaries without needing to remember exact syntax or jump between tabs.

Thatโ€™s how Cellmate AI started.


๐Ÿ”ง Building with Bolt

Once I committed to the idea, I had around 15 days left in the World's Largest Hackathon presented by Bolt. To move quickly, I relied on Bolt.new to scaffold most of the application โ€” from UI components to basic functionality.

Almost every major feature started with a Bolt prompt.

Some initial examples:

  • "Create a React spreadsheet grid with editable cells"
  • "Add a formula bar which will for now contain the cell value"
  • "Add a toolbar with basic formatting like color, bg, alignments"
  • "Add CSV import/export support"

Bolt helped me move fast, especially when I broke down prompts into focused tasks. Larger prompts often generated bloated or buggy code, so I kept things small and stitched the parts together manually.

When Bolt-generated output broke existing logic or styling, I cleaned it up myself. I avoided over-engineering and left out anything that wasn't essential.


๐Ÿง  Prompt Strategy and Workflow

My workflow eventually settled into this loop:

  1. Write a clear, single-purpose prompt
  2. Let Bolt generate a scaffold
  3. Test it immediately
  4. Patch or rewrite the pieces that broke
  5. Move to the next task

By keeping each step tight, I avoided the usual AI-overhead and kept things predictable. This approach worked well โ€” especially when combining AI-generated logic with my own cleanup.


๐Ÿ“ธ What the App Does

Cellmate AI is a lightweight spreadsheet app with built-in AI support โ€” designed to make working with data faster and less manual.

Hereโ€™s what it currently supports:

  • Formula generation from plain text
    Type something like "Sum column B if column C is complete" and it returns a working =SUMIF() formula.

  • Sheet-level changes via prompt
    You can ask it to delete rows, add columns, or clean up sections without touching any menu.

  • Natural language insights
    Ask questions like "Which product had the highest revenue?" or "How many users signed up last week?" โ€” and it gives you answers based on the data in the sheet.

  • Auto-generated summary reports
    One prompt can generate a full summary of the sheet contents.

  • CSV import/export
    Quickly upload or download data.

  • Supabase integration for persistence and auth
    User sessions and sheet data are synced using Supabase โ€” so it works across devices.


๐Ÿงฐ Tech Stack

  • Bolt (scaffolding + code generation)
  • React + TypeScript
  • ShadCN
  • TailwindCSS
  • Vite
  • OpenAI (natural language โ†’ formula/insight)
  • Supabase (auth + database)
  • Hosted on Netlify

โœ… Whatโ€™s Working

  • Spreadsheet grid with editable cells
  • Formula generation from plain text
  • Sheet-level structural changes via prompt
  • Insights and summary report generation
  • CSV import/export
  • User login and data sync via Supabase

Here's a demo video describing the current stage (0.75x might help):


๐Ÿž Whatโ€™s Missing (for now)

  • No multi-sheet/tab support
  • No real-time collaboration
  • No AI-generated charting or visualization tools
  • Some UX rough edges in prompt result placement

๐Ÿ’ก What I Learned

  • Building from a real pain point made it easier to stay focused.
  • Bolt can be used beyond prototyping; it landed some great features by providing clear-cut instructions.
  • Prompt clarity mattered more than prompt length โ€” vague requests broke things quickly. (Thanks to revert/undo option)
  • I didnโ€™t try to do everything, and it helped me finish a foundational MVP.

๐Ÿ”œ Whatโ€™s Next

  • Support for Excel file uploads
  • Summary dashboards and report saving
  • Sharing and collaboration features
  • Possibly releasing a public version with pricing

Thanks to Bolt, DEV, and the hackathon team โ€” the pressure helped me shift gears and build something that I'm happy about.

Try it out: https://cellmateai.xyz

Questions, feedback, bugs? Happy to hear them.

Top comments (6)

Collapse
 
dotallio profile image
Dotallio

I love how you turned the repetitive spreadsheet grind into something so much smarter with natural language prompts. Do you think you'll expand Cellmate AI into handling more general workflow automations beyond sheets?

Collapse
 
hzahar profile image
Hassan Zahar Rifat

Thanks, Dotallio! I think I would eventually. But for now, my focus is on making everything inside the sheet as automatic as possible.

Collapse
 
ziaul_kabirrana_ed9ac707 profile image
Ziaul Kabir

This is great. I think it should be made production-ready, as it would be very helpful for many users.

Collapse
 
hzahar profile image
Hassan Zahar Rifat

Thanks, Kabir bhai. Working towards that.

Collapse
 
nathan_tarbert profile image
Nathan Tarbert

i've enjoyed all of the research you've put into this project - it adds up
you ever feel like building for yourself gets you more excited about the finish line than building for someone else

Collapse
 
hzahar profile image
Hassan Zahar Rifat

Appreciate that, Nathan. Agreed, itโ€™s a special kind of motivation.