Sofia FAQs


What is Sofia?  

Sofia is a conversational AI assistant available on the Suvoda Platform. Sofia helps users access their clinical trial data more effectively through an intuitive chat interface.  


How does Sofia work?

Simply pull up the chat interface and ask Sofia a question. Sofia will try to answer the question in one of two ways:  

  1. Map the question to an existing question in the question bank. The question bank is a collection of questions and corresponding instructions on how to answer those questions. This makes up the teaching framework that allows us to continually expand Sofia’s knowledge over time in a safe, controlled, and predictable way.
  2. Search available reference materials such as the Suvoda IRT user manuals.  

 

What are the key requirements for Sofia?

Reliability  

Suvoda has also invested heavily in building out Sofia with the proper data verification steps in place so that Sofia does not make up data and is consistent in how questions are answered.  

Teachability  

Sofia only answers questions that Sofia has been taught to answer. Training and knowledge growth is directed by Suvoda. Because Suvoda instructs Sofia in advance on how to answer each question using a set of deterministic steps, Suvoda is also able to test Sofia as part of the Suvoda Software Development Lifecycle (SDLC) process.

Security & Data Privacy 

Sofia will provide users with access to data based on the IRT roles and permissions matrix – in short, users will have access to the same data they have access to through the IRT application. The LLM agent that Sofia relies on will have access only to the application data needed to answer the user’s question. Such LLM agent has a zero-day retention policy, and the data is not used to train public models. 

 

How can we be sure we can trust Sofia?  

Each Sofia question is tested following Suvoda’s SDLC process, which includes both positive and negative flow testing. Suvoda’s SDLC also involves testing in two lower-level testing environments across multiple datasets, so each question is tested repeatedly with variable trial data for both accuracy and consistency.  

At a high level, the steps operate something like this: 

  1. John Smith asks a question  
  2. If needed, Sofia will clarify with the user to gather more information to answer the question 
  3. Sofia maps that question to the question bank  
  4. Sofia follows the pre-defined expert instructions to answer the matched question  
  5. Suvoda validates the instructions have been followed, and data has been retrieved and processed from the application when answering the question  
  6. The question is answered  

As with any natural conversation it is possible for Sofia to misunderstand the question so that Sofia’s answer may not map to the user’s intended query, at which point the user can clarify intent with Sofia. It is up to the user to evaluate the response and confirm it matches the expected answer. However, Sofia does not make up data. 

 

Who can use Sofia?  

Access to Sofia will be managed through the already existing roles and permission matrix in the Suvoda Platform.  

 

How does Sofia replace the work people currently do?  

Sofia is meant to assist how people do work, not replace those subject matter experts. Our goal is to help users access the data they need to perform their day-to-day responsibilities in a more effective way. 

Rather than requiring multiple clicks to access data, Sofia can provide a direct link to the targeted space. 

Instead of searching through a dense 50-page reference guide, Sofia can identify and display the exact section of the user manual that answers the user’s question.  

What used to take multiple steps to filter report data and aggregate targeted information can be answered through a question.  

 

Will this data be used to train public models? 

No, while Suvoda does maintain a full trail of the user chat history outside of the LLM, the LLM agent used has a zero-day retention policy and only retains data long enough to engage with and answer the user’s questions.  

 

Who has access to the user question history?  

The user chat history is accessible by Suvoda administrative users. Each user interacting with Sofia will only see their conversation history with Sofia.  

 

Does Sofia support other languages?  

Yes! Sofia can respond to users in the same language as the question.  

 

What questions can Sofia answer? 

Sofia can answer over 100 questions on over 25 topics with plans to increase knowledge over time as we continue to teach Sofia by continually adding questions and instructions. 

Example questions that can be answered are:  

“What is the status of drug unit 1345?”  

“What is the site inventory at site 101? And can you make a bar chart of this data?”  

“What are the different dose statuses and what do they mean?”  

One firm restriction is that Sofia can only answer questions about the clinical trial data the user has access to. For example, Sofia cannot answer questions about sports.  

 

Can Sofia take action on my behalf?  

No, Sofia is currently limited to retrieving data, providing quick access navigation links (when applicable), supporting data visualization, and searching available reference materials such as user manuals.  

Learn more about the Sofia early adopter program

Find out if your clinical trial is a good fit for our early adopter program. Fill out the form below, and we’ll set up a meeting to discuss the program.