Migrating My Vibe-Coded Ranking Monitoring System to Gemini Spark
This article was automatically translated from Japanese by AI. Read the original Japanese version →
Table of Contents
If you've been following this blog for a while, you probably know I previously built a system using Vibe Coding.
In short, it's a system that periodically monitors rankings on the Meta Quest Store, Steam, and the like, and automatically notifies me when new games enter the charts (i.e., games gaining buzz right now). Check out the earlier post for details.
This time, I migrated that system to Gemini Spark. Even though the original was Vibe-Coded, I went out of my way to turn existing code into a prompt and run it on a schedule with Gemini Spark.
In this post, I'll walk through the steps and the reasoning behind the move.
Writing the Prompt
First, I exported the existing system's historical data into a CSV. Next, I loaded the system's repository into Gemini inside Antigravity and gave it the following instructions to generate the prompt. Once you put the generated prompt into Gemini Spark, it handles the rest nicely.
Instructions
I want to port the implementation in this repository to Gemini Spark, so please come up with a prompt for it.
- Store and persist values like those currently in S3 files or YAML in a spreadsheet instead.
- Reference AWS S3 as needed to reflect the existing details in the prompt.
- Since HTML parsing when fetching website info will be left to the AI, don't specify granular implementation details.
- Add "why this title entered the rankings right now" to the output—for example, a sale discount, a major update, etc. Educated guesses are fine.
The Resulting Prompt
Role & Objective
You are a "Multi-Platform Popular Game & Content New-Entry Monitoring Agent." Crawl the specified websites to fetch the latest rankings and update information, cross-reference them against the Google Spreadsheet "Game Ranking Monitoring History Master," and detect, record, and report newly ranked (or updated) titles.
1. Spreadsheet Setup (Initial Run or If Sheets Do Not Exist)
If the linked spreadsheet "Game Ranking Monitoring History Master" does not exist, create a new one based on the attached CSV file (ranking_history.csv).
- Tab (Sheet) Structure: Split sheets by platform (
meta-quest,steam). - Header Columns
- Game ID
- Title
- Best Rank
- First Seen
- Last Seen
- URL
- Rank-in Reason
- Populate each corresponding tab with the data from the attached CSV, freeze the first row, and enable filters. *If the spreadsheet already exists, skip this initialization step and use the existing sheets.
2. Target Platforms & Fetch Criteria
Visit each of the following pages and retrieve the top titles' information:
- meta-quest:
https://queststoredb.com/charts/?period_days=7&pricing=paid&release_days=(Top 50) - steam:
https://store.steampowered.com/search/?vrsupport=402&filter=topsellers(Top 50)
3. Cross-Referencing, Inference & Updating the Spreadsheet
Cross-reference the latest data against each platform's sheet.
- New Entry Detection
- Treat any title whose
Game ID(or URL) is not yet registered in the sheet as a "new entry."
- Treat any title whose
- Inferring the Rank-In Reason (New Entries)
- Based on information on the webpage (sales/discount rate, major updates, new release, going free-to-play, etc.) and recent activity around the title, infer and analyze "why this title entered the rankings right now."
- Writing to the Sheet
- New entries: Append a row to the bottom of the sheet and record the inferred reason in the
Rank-in Reasoncolumn. - Existing entries: Update
Last Seento today's date, and if the current rank is higher thanBest Rank, updateBest Rankas well.
- New entries: Append a row to the bottom of the sheet and record the inferred reason in the
4. Report Output
When new entries are detected, create a clean, well-organized new-entry notification report.
- Completely omit any platform that had no new entries from the notification/report.
- If there were no new entries on any platform, report that concisely in a single line.
Why I Migrated
Honestly, there isn't much more to explain about the system or the prompt itself, so let me explain why I decided to migrate.
The first reason is that while experimenting with Gemini Spark, I got a strong hunch that it could handle way more functionality on its own than I'd expected—scheduled execution, outputting files to Google Drive, browsing the web, searching, and so on. If it can do all that, building a system like this is completely feasible. Above all, being able to use Google Sheets makes it more than capable as both an output destination and a lightweight database.
The second reason is receiving completion notifications directly in the Gemini app. With the previous system, completion reports were delivered as emails via Feedly. Because they went through AWS SNS email, formatting options were limited and it was hard to make them look nice. Viewing them in the Gemini app lets me display anything Markdown supports cleanly, clicking links is smooth, and if I have follow-up questions I can just ask Gemini right there.
The third reason is extensibility—or rather, how easy it is to modify. The previous system was already pretty easy to update since I could request features via Jules and merge PRs on GitHub, but with Gemini Spark it's even simpler: I just ask directly in the Gemini app. Not having to manage any cloud infrastructure at all is a huge breakthrough.
In the prompt above, I casually added why this title entered the rankings right now to the report, and it handled that effortlessly too. I have no idea how it researches and organizes that under the hood, but so far the reasons it gives seem pretty plausible. Having an LLM or AI agent breeze through tasks that would be a pain to implement seriously is what makes them so impressive. Tweaking the report layout and formatting takes just a few minutes.
Using a spreadsheet as a database is also super convenient—browsing historical data is way easier on the eyes now. Previously I was staring at JSON files in S3...
Final Thoughts
The migration went way smoother than I expected, and it's been really convenient to use. Sure, it feels a bit extravagant with token usage, but since I'm subscribed to Google AI Pro, it's effectively free.