I was building allpub.co (a smart cross-platform publishing tool) when I hit a problem that code couldn't solve: which LLM should I actually use?
The 3 AM Problem
It's late. I need AI-powered title generation, SEO extraction, and summaries for AllPub. Simple, right?
Not even close.
Should I use GPT-4? Great quality but insanely expensive. Claude 3 Haiku? Cheap but will it work well enough? Mistral? Llama? Deepseek? Each one has different tradeoffs. Speed versus cost. Quality versus infrastructure headaches.
Three hours later I've got 47 browser tabs open and zero clarity.
The frustrating part? I'm a developer. I can code anything. But I have no idea which LLM is actually right for my use case. And I'm betting thousands of other builders feel the exact same way.
The Spreadsheet That Changed Everything
So I built a decision matrix. What's my actual use case? What tradeoffs can I accept? How many requests per month? Do I need self-hosted or API-based?
Once I answered those questions, Claude 3 Haiku became obvious. Not because it's the best model overall, but because it's the best model for AllPub's constraints.
That's when it clicked: what if there was a tool that did this for everyone?
Enter ArchitectGBT
Here's what I built:
architectgbt.com
You describe your project in plain English. "I need a chatbot that handles 1000 daily conversations with super low latency."
It analyze your constraints and recommend the best models ranked by speed, cost, and accuracy for YOUR use case specifically.
You see pricing breakdown and estimated monthly costs so you know what you're paying for.
You get code templates to start building immediately.
No spreadsheets. No guessing. No vendor bias. Just recommendations that make sense for your project.
Why This Matters
The average engineer wastes 3 to 5 hours picking an LLM. But the real cost is picking wrong and overspending 30 to 50 percent every single month.
I've already heard from early users. One team was using GPT-4 for summarization when Claude Haiku would've worked fine. That's 2,400 dollars a month they could have saved.
ArchitectGBT cuts through the noise. You get the right recommendation for your specific situation, not what's trending on Twitter.
What's Next
architectgbt is in BETA right now and we need your feedback. Here's what I am thinking to build:
- Benchmarking suite so you can test your actual prompts across models.
- Fine-tuning advisor to help you decide between self-hosted and API-based.
- Team dashboard so your whole organization is aligned on model choices.
- Budget alerts before you overspend.
- Model routing that automatically picks the cheapest model that works for each request.
Try It
Go to architectgbt and describe a real project you're working on.
Tell us if the recommendations are spot on or if we're way off.
Join the **waitlist **for the full features. Early testers get discounts.
That's it. The boring stuff works. The expensive stuff is expensive. Your constraints actually matter.
If you've ever wasted a few hours picking an LLM, this is for you.
architectgbt.com
Feedback welcome. Twitter or reply here.
Top comments (0)