How can I safely keep developing my Firebase + Vertex AI projects without huge costs?

General
,
1

Hi everyone,

I’m building two projects to help the poker community:

1. AggroFish Analyzer (Firebase + Vertex AI app): players can input their hands and get AI-powered GTO & exploitative analysis. They can also save hands by date, download PDF reviews, and track their learning progress.

2. A mindset-focused website: a space for poker players (especially over 40 and beginners) to learn stoicism, discipline, and mindset tools. On this site, I want to include an “Ask the Advisor” app, where people can ask questions about strategy, mindset, and improvement and get helpful answers powered by AI.

I started this journey with zero coding experience. I’m not a professional developer, but I’ve been learning step by step using trial and error, tutorials, ChatGPT, and Gemini. My dream is to give players who don’t have money for expensive coaches a way to improve both their technical skills and mindset for free or at a very low cost.

The problem: in September, I accidentally generated a very high bill (~$2,000+) from Vertex AI because of repeated Cloud Run requests to Gemini 2.5 Pro. I’ve since removed the triggers and I’m working with billing support trying to sort out this but now I’m honestly scared to keep working. I want to build the “Ask the Advisor” app, but I’m afraid of testing it and being charged a lot again.

:backhand_index_pointing_right: My questions:

1. How can I test safely in Firebase/Vertex without risking huge charges again?

2. Is there a way to simulate AI responses locally or run a cheaper “dev mode” until I’m ready for production?

3. What quotas or budgets should I set to stop runaway charges?

4. What’s the best way for a solo learner like me to control costs but still make progress?

This project means a lot to me. Before, I was only writing quotes, but now I’m building something real. I don’t want to give up or let my projects die — I just need to learn how to build in a safe way.

Thanks in advance for any advice :folded_hands:

— Victor

2

I personally found that using Vertex AI was like being given a beginner’s pack of cigarettes: they make everything so easy with free credits that you don’t learn how to do anything else. Most if not all of the Google Cloud Skill boost classes teach everything AI or ML on Vertex AI. Shift your development off Vertex AI and learn to use the official Google GenAI SDK or Hugging Face libraries. Switch all your development and testing requests to Gemini 2.5 Flash-Lite its so cheap I have a hard time even seeing a bill. Set up Quotas and Bugets. I have a $.50 cent and$1.00 budget alert. Also learn all about Caching.

3

Agree with above post - I’m not sure you need the Vertex AI for dev/testing purposes, possibly not for production either, depending on your use cases. The Gemini Developer API has generous free limits, and you can set the model used in Genkit (the AI can do this for you). For example, in developing/testing my app, I’m using Gemini 1.5 flash and I’m yet to incur any costs. Again, this really depends on what you need from a model.

Gemini 2.5 pro is quite an expensive model. For context, Grok 4 Fast is as capable as Gemini 2.5 pro on many tasks, but is x20 cheaper for output tokens, huge difference. Have you done a cost analysis of what a typical user would incur using your app (based on input and output tokens)?

I spent quite a bit of time investigating my own AI API costs before building the app to see what’s feasible and whether the app still held water from a costs/revenue perspective.