SheetSentinel

Your AI spreadsheet companion that catches errors before your boss does.

Comment

The Moment That Inspired SheetSentinel

It was end of month. I had spent hours putting together our monthly report, pulling numbers, building summaries, double-checking totals. I was proud of it. I hit send.

Five minutes later, my boss replied.

She had spotted an error I missed. A wrong total. Right there in the report I had just confidently sent to her.

That moment of embarrassment stuck with me. Not because I made a mistake (everyone does) but because I realised there was no safety net. No second pair of eyes. Just me, a spreadsheet, and the hope that everything added up correctly.

I knew I wasn't alone. Every finance analyst, operations manager, and team lead who builds monthly reports knows this feeling. You work carefully, but spreadsheets are unforgiving. A single wrong value can cascade into multiple errors across a report.

That's why I built SheetSentinel.

What I Learned

Building SheetSentinel taught me that the hardest part of AI development isn't the AI. It's the data pipeline. Getting clean, structured data from a real spreadsheet into a format the AI can reason about took more iteration than the analysis itself.

I also learned that prompt engineering is everything. The difference between an AI that just checks numbers and one that understands business context comes down to how precisely you instruct it. Teaching the model to reason about meaning, not just values, was the key breakthrough.

Finally, I learned that the best user experience is the one that requires the least change in behaviour. Users shouldn't have to go somewhere new to get help. The help should come to them, right inside the tool they're already using.

How I Built It

SheetSentinel has two layers:

The AI Backend, built on Airia

An Airia agent powered by Claude Haiku 4.5 receives spreadsheet data in CSV format and analyses it using pattern recognition and business logic reasoning. No hardcoded rules. The AI infers what each column means and applies contextual judgement.

The agent flow:

  1. Python node receives and cleans the raw data
  2. Claude Haiku 4.5 analyses it against a carefully engineered audit prompt
  3. Python node formats the findings for display

The Frontend, a Google Workspace Add-on

A Google Apps Script add-on that:

  • Reads the active sheet directly via the Sheets API
  • Converts it to CSV and sends it to the Airia agent
  • Displays severity-ranked results in a clean sidebar

The result is a companion that lives inside every Google Sheet, always one click away.

Challenges I Faced

Google Sheets doesn't expose its data to browser extensions. My first approach was to use the Airia Chrome extension to read the page. Google Sheets renders as a canvas application and there is no readable text on the page. I had to pivot completely to a Google Apps Script approach, which reads data directly from the Sheets API.

Parsing unstructured AI responses. The AI returns rich, formatted text. Extracting structured flags from that output required careful regex parsing and fallback logic to handle variations in the response format.

Making the AI understand context, not just numbers. Early versions flagged too many false positives. Teaching the model to distinguish between a genuinely suspicious value and an intentional business decision required multiple iterations of the audit prompt.

Building under time pressure. This entire project was built in under 24 hours. Every technical decision had to balance quality against the deadline.

How to Use SheetSentinel

Option 1 — Try it instantly (recommended)

  1. Open the shared Google Sheet in the "Try it out" link
  2. Click Extensions > Apps Script
  3. Go to Deploy > Test Deployments > Google Workspace Add-on > Install
  4. Go back to the sheet and refresh the page
  5. Click SheetSentinel > Audit This Sheet
  6. Sidebar opens, click Audit This Sheet
  7. Results appear in seconds

The sheet already contains test data with intentional errors so you can see SheetSentinel in action immediately.

Option 2 — Use your own sheet

  1. Follow steps 1 to 4 above to install
  2. Open any Google Sheet with your own data
  3. Click SheetSentinel > Audit This Sheet
  4. Click Audit This Sheet in the sidebar
  5. Review your flagged issues

Option 3 — Use your own Airia API key

  1. Sign up free at airia.ai
  2. Create a project and generate an API key
  3. In Apps Script open Code.gs
  4. Replace AIRIA_API_KEY with your own key
  5. Save and redeploy

What to expect

Each audit returns:

  • Summary — one sentence overall assessment of your data quality
  • Issues Found — total count of anomalies detected
  • Flagged Issues — each flag includes:
    • Severity level (HIGH, MEDIUM, LOW)
    • Plain English explanation of the problem
    • What was found vs what was expected
    • A specific suggestion for what to investigate
    • Healthy Areas — confirmation of what looks correct

Coming soon

SheetSentinel is currently available as a developer test deployment. The next steps for the project are:

  • Google Workspace Marketplace listing — one-click install for anyone, no developer setup required
  • Real-time monitoring — automatic auditing as you type
  • Slack and email notifications — get alerted when critical errors are detected in your reports
  • Historical comparison — flag changes across monthly reports automatically
  • Microsoft Excel Online support — bringing SheetSentinel to the full enterprise spreadsheet ecosystem
  • Per-user API key management — each user connects their own Airia account for full data privacy

Note: The test deployment uses the developer Airia account. For production use, replace with your own API key from airia.ai.

The Bigger Vision

SheetSentinel as submitted is the foundation. The real vision is a companion that monitors your spreadsheet as you type, flagging issues the moment they occur, before you even finish the row.

No more end-of-month panic. No more embarrassing emails from your boss. Just quiet, intelligent confidence that your numbers are right.

Because everyone deserves a second pair of eyes.

Built With

Share this project:

Updates