CataractBot: An LLM-Powered Experts-in-the-Loop Chatbot for Cataract Patients

[@techguybiswa]

Really really cool! It would help so many people at such a large scale! Absolutely loved the vernacular language support as well!

[@kbjagadeesh]

Well made video. Very informative and useful.

[@aloktripathi27]

I like this approach

[@anirudhkanneganti1]

I am more interested in how the feedback loop is working in correcting the bot’s answers and whether each of the response is sent for expert verification. To what extent the information is reliable?

[@laxmanshankar867]

The doctor seems to be correcting the bot's reply which was possibly already read by the patient. Isn't this risky especially in healthcare?

[@mohitjain088]

Based on the ophthalmologist's ask, we followed such a protocol, because of two reasons: (1) in 9 out of 10 cases, they found the bot to output a correct response, and (2) the ophthalmologists thought that there is a higher risk of patients getting incorrect information from other sources (like online or friends/relatives). In another deployment of experts-in-the-loop bot for Oncology use case, the oncologists wanted their patients to only receive expert-verified responses. The system is configurable to take care of that as well.

[@archak737]

Interesting. How is it offering a differentiating advantage over directly chatting with the doctor which many doctors offer? Each of bot response has to be anyways verified by the doc in this bot.

[@mohitjain088]

Even for doctors who are available to chat directly with their patients, such a system could potentially reduce their workload.

[@archak737]

Makes sense. I was imagining a "should-I-get-tested-for-covid-or-not" type of fully automated if-else condition based bot.

[@miraculixxs]

How do you ensure that *every* response by the bot is accurate?

[@mohitjain088]

By having experts, such as doctors and patient coordinators, verify every response from the chatbot.