COLO COMPANION

"good work"

"Idea looks nice and good presentation, but ultimately the output of Vectara (text variable) seems dangling and not use in the output. "

"- Good, mission-driven scenario that the application is looking to address. The presentation was good, but I would have liked to see a summary, and perhaps some next steps or thoughts on how you can make this available to people. - There is a good focus on the human element of the application. Very important in this domain. But it did not let me ask a follow up question to the bot. - Clean, well structured repo with a good readme. "

"It looks like the summary from Vectara is then being fed into Anyscale llama2, instead of the search results. Maybe data wasn't loaded into Vectara? I could only find a single PDF in, which didn't have much information in it. As a result, this hallucinates quite a bit, citing sources that don't exist. I asked a few questions in the demo and the cited sources sent me to https://www.mayoclinic.org/diseases-conditions/colorectal-cancer/symptoms-causes/syc-20363775 and https://www.cancer.org/cancer/colon-rectal-cancer/causes-risk-factors.html, neither of which exists. I think the submission would be much more compelling if, instead of taking the Vectara summary, it took the list of results from Vectara and then sent them to Anyscale. This would allow for many of the benefits of retrieval augmented generation, including proper citation citing, which is critical for a use case like a medical chatbot"