I built an AI Gateway, despite having no technical background. Here's where I'm at.
Hello, everyone.
I'm a solo founder based in Argentina, and I created neuralrouting.io. It's an LLM router designed to evaluate prompt complexity and select the most cost-effective model.
Instead of sending everything to GPT-4, it routes simpler queries to cheaper models and caches similar requests so you don’t pay twice.
I also built a Shadow Engine that runs cheaper models in parallel to benchmark quality, filter PII, rate limit, and detect agent loops. It’s still early and only supports OpenAI and Groq right now.
No one’s really using it yet, so I’m looking for devs or small teams working with LLMs who want to break it, give feedback, and tell me what’s missing.
If you wanna try it out or ask questions, check it out at neuralrouting.io or just reply here!
Would love to hear your thoughts.
Building an AI Gateway as a solo founder is a massive technical feat, sheldor. Most teams are bleeding money by over-powering simple prompts; your "Shadow Engine" logic for benchmarking cheaper models in parallel is exactly the kind of cost-optimization that turns a "vibe" project into a sustainable business.
I’m currently running Tokyo Lore, a project in Tokyo that highlights high-utility logic and first-principles builders like you. Since you're tackling the expensive problem of agent loops and PII filtering, entering NeuralRouting could be the perfect way to get your gateway in front of serious developers while your odds are at their absolute peak.