RAG + LLMs
Explain and Evaluate RAG/LLMs
Howso connects directly to the source data and operates as the ground truth in a RAG or LLM system.
Go beyond explaining AI. Prove it.
Step #1
LLM is fine-tuned or used in a RAG process.
Step #2
The data used for LLM fine tuning or in the RAG process is loaded into Howso.
Step #3
As each LLM query is made, Howso compares the result to the original data.
Step #4
Howso calculates the probability that the prediction was made using your data versus pretrained weights, giving exact attribution for what parts appear to be inferred from your data.
Step #5
Additional debugging can be performed in areas of hallucination. For example, if Howso reports that the documents used in inference appear to be different than what was used in a RAG process, it is more likely that the results are a hallucination. This can lead to automatic remediation or raise an issue for additional system tuning.
Join the Howso Design Partner Program.
Partner with us to usher in the era of AI explainability. Get access to our latest products.