How I Turned an Ugly Spreadsheet into an AI Assisted App with Antigravity

Source: Dev.to

From a Painful CSV to a Fully Deployed Cloud Run App

I needed to review a massive spreadsheet of conference talk submissions. Staring at tiny cells was exhausting, so I decided to build TalkScout, a UI‑driven review tool, and use Google Antigravity (Google’s AI‑powered coding agent) to handle the heavy lifting.

Using Meta‑Prompting to Kick‑Start the Project

Instead of writing code first, I chatted with Gemini 3. I described my problem in plain language, and Gemini 3 acted as an architect, turning my “brain dump” into a polished technical specification. This spec defined the component hierarchy and data model, which I then fed directly into Antigravity.

Let Antigravity Build the UI

With the spec, Antigravity generated a React app that lets me:

• Upload the CSV containing all conference talks.
• View a dashboard showing the status of each submission.
• Review abstracts and demo plans in a high‑contrast, distraction‑free interface.

[Image: TalkScout Dashboard (synthetic data)]

Fixing a React Hydration Error

During the first run I hit a React hydration mismatch error. I simply pasted the error message back to Antigravity, and the coding agent identified the DOM inconsistency and fixed it within minutes.

Adding AI‑Powered Social‑Media Assessment

To avoid personal bias and gauge real‑world interest, I added a button that triggers an AI Assessment. Using Google Search Grounding, the AI searches Reddit, X (Twitter), and LinkedIn for developer signals, providing insights based on actual audience mindshare.

[Image: TalkScout submission review page with AI social media analysis]

Calibrating the “Strict” Reviewer

The initial AI reviewer was overly generous, rewarding any submission with trendy buzzwords. I applied few‑shot prompting to calibrate it:

• Marketing Fluff Penalty – dock points for marketing‑style language.
• Length Cap – cap scores at 2 for overly short submissions.
• War Stories Bonus – increase rating for concrete learnings and real‑world stories.

After a handful of examples, the reviewer aligned with my evaluation criteria.

Refactoring for Batch Mode

Reviewing each talk individually was time‑consuming. I asked Antigravity to refactor the backend to process submissions in batch mode. TalkScout now processes the entire pool in the background, populating an “AI Draft” column with insights while I grab a coffee, allowing me to focus on final decisions.

One‑Click Deployment to Cloud Run

Satisfied with the tool, I wanted to share it with other reviewers. I simply instructed Antigravity to deploy the app. It automatically:

1. Detected my Google Cloud project ID.
2. Containerized the application.
3. Generated the necessary gcloud commands.
4. Deployed the service to Cloud Run.

Within minutes I had a public URL to distribute to the review team.

---

Building TalkScout demonstrated how AI‑assisted coding (vibe coding) can turn a tedious spreadsheet into a polished, collaborative app with minimal manual coding. If you’ve solved a problem using vibe coding, I’d love to hear about it. And if you haven’t tried Antigravity yet, give it a spin and see how easily you can deploy apps to Cloud Run.