Demos
Building a Better way to Practice Functional & Integrative Medicine
Functional medicine emphasizes a personalized, systems-based methodology focused on root cause analysis, prevention, and long-term health improvement. By understanding patients’ unique genetic, biochemical, and lifestyle factors, practitioners can develop more tailored, sustainable healthcare solutions. All while being a vital partner to conventional medicine. However, mainstream medical guidelines often lack the depth needed to fully support functional medicine practices.
FunctionalMind™ bridges this gap by providing practitioners with precise, evidence-backed insights from the latest research. With the advancement of Generative AI, the company is rolling out a novel AI solution that is comprehensive, trustworthy, and personalized for each patient story. It replaces manual research work that could take days – which healthcare providers often just don’t have the time to conduct. Functional medicine often deals with intricate, multi-system issues. Practitioners can find it challenging to develop personalized care plans that consider root causes and interconnected health factors. This new solution provides them with the tools necessary for advanced clinical decision-making, derived from current clinical guidelines and research.
This new AI solution is powered by John Snow Labs, the award-winning Healthcare AI company and the world’s leading provider of Medical Language Models. John Snow Labs’ Generative AI platform can access, understand, and apply the latest evidence-based research from the most authoritative knowledge bases. By partnering with John Snow Labs, FunctionalMind™ provides functional and integrative medicine practitioners with a reliable, thorough and intuitive academic research and clinical decision support chatbot resource designed to keep them at the cutting edge of patient care. Generative AI responses powered by John Snow Labs understand medical jargon and intent, dynamically select which knowledge bases to consult for each question, take into account the context of a patient’s history, and summarize an answer along with cited and validated references.
What makes this solution different from general-purpose Generative AI models and tools is the accuracy, tuned specifically for medical advisory tasks; the fresh and thorough knowledge base on functional medicine; the security and privacy controls; and the ease of use for medical professionals.
“Clinical questions that could take hours or days of research can now be answered in under a minute, at the same level of trust.”
Key Capabilities
- Empowering Evidence-Based Practice: FunctionalMind™ supports physicians in learning and applying evidence-based guidelines in functional medicine, but also in staying up-to-date with relevant research.
- Research Planning & Reasoning: John Snow Labs developed a custom AI Agent so that the system goes beyond a chatbot to conduct planning, reasoning, and harmonization of answers from multiple knowledge bases on the fly.
- Daily Update of Knowledge Bases: The system integrates extensive medical knowledge bases from John Snow Labs alongside proprietary databases, ensuring up-to-date, relevant information.
- Tuned for Healthcare Providers: Data scientists and medical doctors from both teams worked closely together to refine the accuracy, safety, relevance, and writing style of the AI’s answers and summaries.
Building a Robust Solution with John Snow Labs
Leveraging John Snow Labs’ software, language models, and expert team enabled the rapid deployment of a high-performance chat platform in just six weeks, and the addition of a Clinical Advisor AI Agent in just six more weeks. The solution includes the following features:
Referenced Responses:
The chat interface provides cited references from knowledge bases with each answer, ensuring reliability and transparency. Each answer includes a list of references used to generate the response but also additional references relevant to the question.
Follow on questions:
Automatically generated questions can be selected to continue the discussion, much like with a colleague, or individual questions can be used to further qualify clinical or research enquiries.
Clinical Advisor Agent:
Unlike generic AI tools, the Agent uniquely combines patient-specific memory, dynamic knowledge base selection, and current clinical guidelines to provide tailored guidance that is tuned specifically for functional and integrative medicine.
Document Upload and Query:
Allows practitioners to upload research papers, clinical documents, and consultation notes, and ask targeted questions about their content.
Adaptive Tone and Style:
Customizes responses for various audiences, allowing clinicians to switch between professional specialties, specific output focus, and patient-friendly tones.
The Clinical Advisor AI Agent
While general-purpose AI tools deliver a chatbot experience, or can retrieve context using RAG for a specific conversation, the FunctionalMind™ solution goes further thanks to the development of an AI Agent that imitates the workflow and thought process of medical doctors. This combines patient-specific memory, dynamic knowledge base selection, and current clinical guidelines to provide tailored guidance – that is tuned specifically for functional and integrative medicine.
The Clinical Advisor Agent leverages multiple Large Language Models (LLMs), specialized tools, and sophisticated prompting techniques to provide comprehensive medical insights. Its workflow is composed of the following phases:
- Dynamic Planning Phase
- The system analyzes user inputs (questions, instructions, or requirements).
- It generates a dynamic, multi-step execution plan scaled to query complexity.
- Each step is structured with specific objectives, contextual requirements, and user considerations, which are remembered from previous conversations.
- Information Retrieval Phase
- Each step’s objectives are transformed into precise tool queries
- The system employs a “Medical Research” tool that interfaces with a vector database containing millions of scientific papers
- Several optimized queries are generated per question, to ensure comprehensive coverage of each question from multiple angles
- Each query retrieves a set of most relevant documents from the databases.
- Analysis Phase
Specialized, domain-specific LLMs process analyse each returned document against:
- Original query objectives (what do we need to know or do next?)
- Provided context (who is this patient?)
- User-specific considerations (who is asking, a general practitioner or specialist?)
- Clinical relevance criteria (use official guidelines only, or also pre-prints?)
- Synthesis Phase
- A more powerful LLM aggregates and synthesizes the analyses from previous steps
- It generates a cohesive, comprehensive response aligned with the original query
- It ensures clinical accuracy and relevance of the final output
The agent supports parallel processing of the Information Retrieval and Analysis phases for faster performance. This architecture enables the Clinical Advisor Agent to provide well-researched, contextually appropriate medical insights while maintaining computational efficiency through parallel processing.
The Multi-tiered LLM architecture balances specificity and comprehensive understanding. This provides both for higher accuracy, since specialized LLMs are fine-tuned to specific tasks, as well as cheaper & faster inference – since each task-specific LLM is smaller.
Dynamic query generation and optimization provides maximum relevance, mitigating some of the traditional limitations of vector databases and RAG systems. Asking a question in different ways enables finding more potentially relevant papers, and separating the Analysis and Synthesis phases enables effectively using a large number of retrieved results, which is often otherwise challenging even for long-context LLMs.
Medical Knowledge Bases: Private, Comprehensive, and Fresh
Another aspect in which this solution differs from alternatives – such as Generative AI Co-Pilots, Search, and Summarization tools – is in the quality of functional medicine content at its disposal. John Snow Labs’ Generative AI Platform comes with a suite of medical knowledge bases that are indexed, enriched, and updated daily, including all public medical research and terminologies. FunctionalMind™ uses the John Snow Labs platform to regularly add a range of additional private knowledge bases to ensure responses are more comprehensive. Users can select access to specific knowledge bases as needed, and can choose which knowledge bases to use for each conversation. Key data sources at launch include:
- PubMed: 30+ million articles covering the breadth of medical research.
- BioRxiv: 40,000+ articles offering pre-peer-reviewed insights into recent advancements.
- MedRxiv: 100,000+ articles focused on health sciences preprints.
- Oregon State University’s Drug-Nutrient Interactions: Extensive data on interactions critical for functional medicine.
- Micronutrient Information Center: Information on essential micronutrients, their benefits, and clinical implications.
- Proprietary Databases: Curated content from patient cases and articles that provide additional context and detail.
By leveraging these specialized knowledge bases, FunctionalMind™ delivers insights uniquely tailored to the functional medicine field, enabling practitioners to answer complex, specific patient questions with evidence-based responses.
Delivering Enterprise-Grade Security, Speed and Scalability
The system is deployed as a dedicated installation of the John Snow Labs’ Generative AI Platform within FunctionalMind’s infrastructure and security perimeter. All LLMs are deployed locally, so that no external API calls are made, and no data is shared outside of that secure and private environment. This ensures that the proprietary content, patient data, conversations, and feedback that the FunctionalMind™ community will provide are not shared with any third party.
The Generative AI platform is deployed on Oracle Cloud Infrastructure (OCI). OCI provides highly optimized Nvidia servers, including unique hardware shapes tuned for serving LLM models and applications. It also provides a broad suite of security and compliance controls which are necessary for a medical application.
The FunctionalMind™ frontend applications run on AWS. API integration allows the team to control and personalize every aspect of the solution, ensuring a seamless, branded experience for its user base.
Generative AI is advancing at lightning speed and there is a long roadmap of innovative features that are planned to benefit the functional medicine community – from automated literature reviews to new modalities of patient data. Expect this strategic partnership to keep advancing the state of the art in Medical Generative AI, while making it more easily usable to more medical professionals every day.
