
How aRCie Works Looking for a more personalized way to search for resources? aRCie is here 24/7, giving you tailored answers using trusted, up-to-date info from experts! You don’t need to double-check the reliability— aRCie has already done that for you! You can ask aRCie questions about specific resources by clicking the Ask aRCie button on an individual record. Try asking aRCie to provide a ten question multiple choice quiz with answers in bold for the following 87 page book "Just In Time Handbook: A Quick Pediatric Reference Guide for Adult Healthcare Providers". You can also ask aRCie broad questions and it will search thousands of resources to provide you a custom answer. Try asking aRCie for a workshop or lecture outline covering X topic for Y audience for Z length of time. This page will explain how aRCie works to provide you personalized answers to any questions you have that are relevant to caring for children in everyday emergency and disasters. The aRCie Chatbot AI (RAG-based) Architecture aRCie's technology is based on a Retrieval-augmented generation (RAG) architecture. This means that it combines the strengths of retrieval-based models with generative models to improve the accuracy and relevance of generated text. When a user asks a question, aRCie uses a type of search called vector-based search, which looks at meaning instead of just keywords. It knows that "car" and "automobile" mean the same. aRCie shows the Large Language Model (LLM) 1-3 articles from the vetted knowledge base of Resource Central, restricting answers to those sources to prevent misinformation. Unlike regular AI setups that pull from both client data and the internet, aRCie strictly controls the information used! Time Delay: Why is aRCie sometimes slower than other AI chatbots?* We're running a lot of rules in the background to make a response accurate and tailored. Because we are A) asking it for vector-based search from the vetted knowledge-base, B) to tailor a response based on the user's role, C) double check to ensure a valid answer - it will be at least 3 times longer than the regular search! We are also focusing on generating responses in plain language. If we run fewer rules, we'll get a response faster, but then we increase the risk of inaccurate information! Choosing between lighting and diamond responses. In response to user feedback, we recently introduced the option to select a lighting or diamond response. Lighting responses complete steps A and B above while diamond responses complete steps A, B, and C above. Extra Notes: Metadata All Resource Central resources are provided to LLM in advance to generate some metadata that helps improve aRCie's search and responses. LLM generates a description, brief description, and MeSH terms for each article. They are used to improve aRCie's search capabilities. The brief description is included along with the article title when aRCie cites a particular resource to help provide additional context to the user when aRCie presents an answer. Accuracy & Misinformation* While generative AI is fast and smart, it can sometimes miss instructions and give incorrect information. aRCie, however, only creates responses based on its own trusted knowledge base. In contrast, ChatGPT, which is supplemented with outside information, can have 1-5% misinformation. Even ChatGPT’s own File Search feature states in its documentation, “File Search augments the Assistant with knowledge from outside its model…” This means you can’t fully control what sources ChatGPT uses to answer questions. Without the ability to control where the information comes from, there’s always a chance for mistakes or misinformation. Test: chatGPT vs. aRCie Can you be 100% sure there’s no misinformation? Try asking some questions yourself and see! Prompt: Create a list of the top 5 critical patients who require immediate evacuation due to disaster conditions

How aRCie Works

Looking for a more personalized way to search for resources? aRCie is here 24/7, giving you tailored answers using trusted, up-to-date info from experts! You don’t need to double-check the reliability— aRCie has already done that for you!

You can ask aRCie questions about specific resources by clicking the Ask aRCie button on an individual record. Try asking aRCie to provide a ten question multiple choice quiz with answers in bold for the following 87 page book "Just In Time Handbook: A Quick Pediatric Reference Guide for Adult Healthcare Providers".

You can also ask aRCie broad questions and it will search thousands of resources to provide you a custom answer. Try asking aRCie for a workshop or lecture outline covering X topic for Y audience for Z length of time.

This page will explain how aRCie works to provide you personalized answers to any questions you have that are relevant to caring for children in everyday emergency and disasters.

The aRCie Chatbot AI (RAG-based) Architecture

aRCie's technology is based on a Retrieval-augmented generation (RAG) architecture. This means that it combines the strengths of retrieval-based models with generative models to improve the accuracy and relevance of generated text.

When a user asks a question, aRCie uses a type of search called vector-based search, which looks at meaning instead of just keywords. It knows that "car" and "automobile" mean the same.

aRCie shows the Large Language Model (LLM) *1-3 articles from the vetted knowledge base of Resource Central, restricting answers to those sources to prevent misinformation. Unlike regular AI setups that pull from both client data and the internet, aRCie strictly controls the information used!

Time Delay: Why is aRCie sometimes slower than other AI chatbots?

We're running a lot of rules in the background to make a response accurate and tailored. Because we are A) asking it for vector-based search from the vetted knowledge-base, B) to tailor a response based on the user's role, C) double check to ensure a valid answer - it will be at least 3 times longer than the regular search! We are also focusing on generating responses in plain language. If we run fewer rules, we'll get a response faster, but then we increase the risk of inaccurate information!

Choosing between lighting and diamond responses.

In response to user feedback, we recently introduced the option to select a lighting or diamond response. Lighting responses complete steps A and B above while diamond responses complete steps A, B, and C above.

Extra Notes: Metadata

All Resource Central resources are provided to LLM in advance to generate some metadata that helps improve aRCie's search and responses.
LLM generates a description, brief description, and MeSH terms for each article. They are used to improve aRCie's search capabilities.
The brief description is included along with the article title when aRCie *cites a particular resource to help provide additional context to the user when aRCie presents an answer.

Accuracy & Misinformation

While generative AI is fast and smart, it can sometimes miss instructions and give incorrect information. aRCie, however, only creates responses based on its own trusted knowledge base. In contrast, ChatGPT, which is supplemented with outside information, can have 1-5% misinformation. Even ChatGPT’s own File Search feature states in its documentation, “File Search augments the Assistant with knowledge from outside its model…” This means you can’t fully control what sources ChatGPT uses to answer questions. Without the ability to control where the information comes from, there’s always a chance for mistakes or misinformation.

Test: chatGPT vs. aRCie

Can you be 100% sure there’s no misinformation? Try asking some questions yourself and see!

Prompt: Create a list of the top 5 critical patients who require immediate evacuation due to disaster conditions