kg-evaluator

Inspiration

knowledge graphs are the core for improving llm reasoning and reducing hallucinations

What it does

scrape wandb customer website, cleans the site, a prompt creates a knowledge graph, evaluator runs prolog to check if the knowledge graph works well

How we built it

asked chatgpt to come up with a skeleton of a kg, created kg creation agent that has tracing in weave, created an evaluator that runs prolog

Challenges we ran into

python prolog library didn't work, had to do system calls. finding a usecase was challanging

Accomplishments that we're proud of

coming up with a usecase of finding customers, integration with prolog

What we learned

how to use weave, marketing usecase,

What's next for kg-evaluator

Built With

• prolog
• weave