MDClone: Setting New Standards for clinical analysis with ADAM
MDClone is a successful startup that provides the ADAMS Platform, an advanced analytics platform designed specifically for medical and clinical personnel. This platform has already brought advanced data analytics capabilities to clinical users, offering a HIPAA-compliant environment for data investigation without compromising patient privacy.
ADAMS Highlights:
HIPAA-compliant environment
Revolutionizing clinical data handling
Uncovering insights that were previously inaccessible
The revolution of Generative AI, particularly in the realm of chatbots, has the potential to further revolutionize tools like ADAMS. Once implemented, it will enable users to interact with ADAMS through a clear chat interface, making the interaction more intuitive. This allow clinical professionals and researchers who might be less familiar with analytical techniques to perform complex queries and data analysis in their own professional language, without the need for writing SQL or Python code.
TensorOps and MDClone: Pioneering Advanced Analytics in Healthcare
Adopting emerging technologies such as Gen-AI can present challenges even for the most cutting-edge startups, primarily because the adoption of new technology is not their core business. Instead, their focus is on integrating these technologies into their products and business models. On the other hand, TensorOps specializes in the adoption of Gen-AI technologies, and we were honored to be selected by MDClone as their partner to accelerate the integration of generative AI into its platform.
Elevating ADAMS Capabilities with GenAI
The collaboration between TensorOps and MDClone embarked on an ambitious project to enhance ADAMS by integrating Large Language Models (LLMs) for a GenAI solution for data analysis, visualization and statistical modeling. This integration has allowed for a more intuitive interaction with the platform, enabling users to conduct complex data analyzes through a chat interface, without the need for any prior programming knowledge.
GenAI Integration:
Chat interface for intuitive interaction
Advanced AI-driven analyses
Code interpreter capabilities
In the context of ADAMS, LLMs were employed to interpret natural language requests and conversations from users, translate these requests into data analysis tasks, and then execute these tasks in a safe Python environment to provide insights back to the users.
The proposed architecture shown on diagram above is built on MDClone’s cloud on Azure to ensure data privacy and compliance. Currently, Azure is the only provider of LLMs, such as GPT models, that are compliant with the strict regulations such as HIPAA.
The current solution centers on the “LLM backend”, developed by the TensorOps team, which is responsible for the following tasks:
Respond to user chat requests and a specific API REST endpoint;
Connect to LLMstudio (LLM Proxy) to request LLM responses from external providers or locally deployed models;
Send “code execution” tasks to the Code Executor Environment and provide the results back to the LLM for further reasoning;
Connect with MDClone’s Data Warehouse to fetch and write data and artifacts such as data visualizations and updated datasets;
TensorOps brings a unique and structured approach to Gen-AI projects including the implementation of its open source framework: LLMstudio.
LLMstudio allows routing, monitoring and logging of LLM calls and for this project we extended its support in LangChain to enable building complex agents and chains. Learn more about it here.
Ensuring Accuracy, Compliance, and Privacy in ADAMS for the Medical Community
One particular challenge that this project faced is the need for very high accurate results. Unlike other systems where errors might lead to embarrassing situations—which are critical enough—in the realm of medical data, compliance and accuracy are non-negotiable. Within the ADAM, Large Language Models (LLMs) undergo rigorous validation using medical datasets and terminologies to ensure high fidelity in their responses, addressing the critical nature of medical information. Moreover, the necessity for HIPAA compliance is imperative, leading to the exploration of ways to utilize LLMs beyond the Azure OpenAI solution. Compliance was assured by Azure OpenAI’s provision of GPT models integrated via LLM Studio to avoid vendor lock-in.
To reinforce accuracy, ADAMS’ chatbot uses finely crafted prompt-engineering techniques adapted to deal with medical datasets, ensuring a secure environment that strictly adheres to privacy laws. This dual focus on privacy and accuracy underscores the commitment to maintaining the integrity and confidentiality of sensitive medical data. Moreover, the proposed architecture ensures that no data, important or not, will be shared to third parties keeping it within MDClone’s cloud boundaries.
Recognizing the diverse operational needs of medical institutions, the enhanced ADAMS platform offers flexibility in deployment. It caters to various preferences by supporting multiple deployment models, including on-premise installations. This approach enables organizations to choose a deployment strategy that aligns with their specific IT infrastructure and privacy requirements, reinforcing the adaptability and versatility of the ADAMS platform.
From Ideation to MVP In Only Three Months
The success of this project is attributed to MDClone's deep understanding of their customer needs and excellent execution capabilities of their R&D. On top of that, TensorOps' structured approach played a crucial role in accelerating the process of moving from ideation to minimal viable product (MVP) in only three months, by providing dedicated frameworks for developing GenAI applications. The project kicked off with a rapid discovery and assessment phase, whose main objective was to analyze the customer's business objectives, understand their existing data and machine learning setup, and preemptively address any potential roadblocks that could pose challenges in subsequent implementation phases.
Over a condensed 3-4 week period, Solution Architects (SAs) and Subject Matter Experts (SMEs) conducted this assessment. The assessment culminated in a concise report that presented key findings and actionable recommendations, complete with estimates for the effort and time required. This successful ML Assessment had set the stage for more extensive engineering engagements, focused on achieving tangible Proof of Concept (PoC) or Minimum Viable Product (MVP) in very short timelines positioning MDClone ahead of its competition.
Customer Feedback
The name of the assistant we built is Adam. Watch the demo here.
“Adam is an expert on statistical analysis techniques and can help you select the right statistical test, apply statistical approaches to your data, and visualize data in dynamic ways with simple chat requests.
Adam works on your data behind your firewalls. Adam can examine original or synthetic data depending on what the user has access to. This means you can keep your data protected whether you are focusing on an internal improvement effort or an external collaboration.”
Conclusion
The partnership between TensorOps and MDClone represents a significant milestone in the application of AI technology in healthcare. By combining MDClone's innovative vision with TensorOps' AI expertise, we have created a platform that not only advances clinical data analysis but also sets a new standard for privacy, accuracy, and user accessibility in healthcare technology. This success story underscores the potential of AI to transform industries by making advanced data analysis more accessible and impactful.