I built this because AI kept messing up my document summaries.
I was generating contract summaries, and the AI would constantly get things wrong, missing party names, wrong dates, invented clauses. I'd retry 3-4 times per document until it finally worked. Annoying and expensive (all those API calls add up).
So I started writing validation rules to auto-reject the bad ones. But hardcoding rules meant redeploying every time I wanted to tweak something. And honestly, sometimes a human could just fix it in 10 seconds instead of me spending an hour writing perfect regexes.
That's why I built Check Your AI:
1. POST your AI response
2. It gets auto-approved/rejected by Lua rules, OR goes to a human reviewer
3.Webhook delivers the result
Tech stack: Elixir/Phoenix. I wanted to learn Elixir :), and LiveView made the real-time review dashboard easy, Whole thing runs on Gigalixir.
I built this because AI kept messing up my document summaries. I was generating contract summaries, and the AI would constantly get things wrong, missing party names, wrong dates, invented clauses. I'd retry 3-4 times per document until it finally worked. Annoying and expensive (all those API calls add up). So I started writing validation rules to auto-reject the bad ones. But hardcoding rules meant redeploying every time I wanted to tweak something. And honestly, sometimes a human could just fix it in 10 seconds instead of me spending an hour writing perfect regexes.
That's why I built Check Your AI:
1. POST your AI response 2. It gets auto-approved/rejected by Lua rules, OR goes to a human reviewer 3.Webhook delivers the result
Tech stack: Elixir/Phoenix. I wanted to learn Elixir :), and LiveView made the real-time review dashboard easy, Whole thing runs on Gigalixir.
https://checkyour.ai
Would love feedback on:
- Does Lua for validation rules make sense, or would you prefer something else?
- Anyone else need human-in-the-loop for AI output?
Hi, the link goes to the reigtration directly. May be you can change to the landing instead.