LET'S TALK

Find out how we can help you tackle your healthcare challenges.

blog 0923- MapAI_ Breast Cancer in _50-blank
BACK TO PERSPECTIVES
DISPLAY TITLE VALUE:

Building a Cutting-Edge Conversational Assistant: Insights From MapAI's Lead Engineer

blog 0923- MapAI_ Breast Cancer in _50-blank

Thomas Fion, the Engineering lead behind our healthcare data analytics assistant, shares insights on how his team is revolutionizing insight generation with generative AI.  

As the volume of healthcare data grows exponentially, so does the potential for powerful and in-depth insights. But it also increases the complexity of sifting through noise to identify what’s relevant. Traditional tools struggle to keep pace with the growing input, leaving teams frustrated by the need for higher levels of expertise to generate meaningful insights, correct inefficient processes, and resolve fragmented outputs.  

See MapAI in Action! _ButtonMapAI™, Komodo Health’s generative AI analytics assistant, offers a solution by integrating advanced GenAI models with our trusted Healthcare Map™ to provide personalized and informative responses to healthcare queries. Embedded within the MapLab ecosystem, MapAI allows users to engage with complex healthcare data through simple, conversational language. It answers questions related to disease landscapes, patient journeys, and HCP/HCO trends by filtering out irrelevant data and focusing on what matters most. The tool also emphasizes transparency, providing clear visibility into both the data and the methodologies behind its insights, ensuring users can rely on accurate, trusted information from our Healthcare Map. 

How We Did It: Building a Conversational Model 
We began by developing an internal AI conversational engine as a service, integrated directly into the Komodo platform and its data services. This engine employed a central intelligence agent to understand the intent behind each question, allowing it to effectively delegate tasks to specialized GenAI agents capable of querying our extensive Healthcare Map. We experimented with several open-source foundational models, including Llama 3.1, Mistral 7B, and Phi-3, to identify the most suitable option for each specific use case. The MapAI engine leverages the appropriate model to orchestrate our APIs and address user intents, automatically ensuring the latest data recency, quality, and tenancy based on user subscriptions. 

Building a Conversational Model _TabletOur API-first approach enables rapid agent development and seamless integration with our platform’s comprehensive set of APIs, which manage user medical definitions, resource libraries, and access to the industry's most extensive source of patient-centric insights. To streamline the deployment and orchestration of these models, we implemented frameworks like LangChain and LangGraph and utilized state-of-the-art GenAI techniques such as multiagent graphs, fine-tuning based on API specifications and numerous sample scenarios, and advanced prompting methods like RAG (retrieval-augmented generation) for inference of user resources.

In building MapAI, our primary goal and challenge were to create two distinct interaction modes: a guided, democratized mode, where users can easily explore and analyze our Healthcare Map while also being able to edit their drafts side by side; and an expert mode, where users can start from a more advanced point in their analysis and jump directly into KPIs or visualizations.

MapAI in Action: Exploring Recent Breast Cancer Trends With MapAI Assistant
Let’s put MapAI in action to look at some recent trends in breast cancer. Rates of breast cancer are rising in patients under 50, and the guidelines for screening have recently been updated for this group. We asked MapAI to gather insights on this cohort: 

Carrousel Image v2 1
Here we see that the tool detects that a colleague already worked on a similar inquiry. We can choose to use that cohort, or make a new one. The tool then automatically computes, using Komodo’s Healthcare Map data, the number of patients in our requested scope. The tool also then offers the creation of resources for further analysis outside of the chat.
Carrousel Image v2 2
Here we see the tool’s ability to quickly generate data visualizations to better understand the functional context of our inquiry. Again we can see MapAI explaining what it captured from my questions, how it translates that into a query for the Healthcare Map, and the visually plotted response. 
Carrousel Image v2 3
Let’s try different kinds of visualization. We can generate two simple bar charts representing our cohort across demographics of interest. 
Carrousel Image v2 4
Let’s try different kinds of visualization. We can generate two simple bar charts representing our cohort across demographics of interest. 
Carrousel Image v2 5-2
We could also go on to ask for more complex visualizations, setting a condition or event as a reference, and then analyzing patient history around these markers, or for more specific timeframes. 


Given the rapid evolution of generative AI, we focused on automating the entire large language model (LLM) lifecycle, enabling the adoption of future foundational models, and orchestrating and training strategies using frameworks like MLflow. In future versions, MapAI will cover all of MapLab’s capabilities, including dataset generation and dashboard creation. MapAI already streamlines workflows and delivers insights across the enterprise, but future updates will introduce multiuser conversations, where shared resources, follow-ups, and team comments will further enhance and accelerate the dashboard creation process.

By harnessing generative AI and prioritizing user experience, MapAI has redefined what's possible in healthcare data analysis, enabling teams to extract valuable insights from complex datasets with unprecedented ease and speed — up to 10 times faster than existing solutions. Tasks that once required extensive technical expertise and months of effort are now accomplished in minutes, using simple, natural language queries that generate actionable insights and visualizations.

As data scientists, we are just beginning to tap into the full potential of this technology. The future promises even greater advancements as we introduce new interaction models and collaborative features. At Komodo Health, we take pride in leading this revolution, empowering healthcare professionals to make data-driven decisions more easily and confidently. 

Learn more about MapAI, or the importance of data quality in new AI tools.

To see more articles like this, follow Komodo Health on X, LinkedIn, or YouTube, and visit Insights on our website.

By providing your email address, you agree to receive marketing communications from Komodo Health. For more information on how we process personal information, please refer to our published Privacy Notice.
SUBSCRIBE
Recent Stories