We hate clicking through Elitigation to verify that the sources cited by a LLM is correct. So we created 4 deterministic layers + 1 LLM layer to verify that our citations are correct and accurate.
These 4 deterministic layers were inspired by how a lawyers checks citations and they are made to fail fast, so that time is not wasted.
Layer 1 checks if the citations exists Layer 2 checks if the content comes from a Whitelist, Graylist or Blacklist Layer 3 checks if the output aligns with the sources Layer 4 checks if the output answers the main query (LLM) Layer 5 checks if the output is correct
With these 5 layers, we are able to catch hallucinations, unreliable answers, misaligned outputs, and nonsensical answers before you have to.
Lastly, we didn't want to change the workflow you are already used to, so our solutions works on claude.ai and chatgpt.com.
Built With
- celery
- claude
- docker
- fastapi
- pgvector
- postgresql
- redis
Log in or sign up for Devpost to join the conversation.