Moca-bot

Inspiration

As we enter the latter half of the 21st century, the global shift toward an aging population is accelerating at an unprecedented pace. By 2050, the proportion of individuals over 60 is projected to reach 22%, fundamentally transforming healthcare priorities. While acute care once dominated resource allocation, the future clearly demands a robust and proactive approach to managing chronic conditions—particularly neurodegenerative disorders and cognitive decline. Access to continuous care, especially in underserved and remote regions, remains a significant barrier. Fragmented systems, limited specialist availability, and burdensome travel requirements exacerbate the challenges faced by older adults. To address this gap, our team is advancing a vision for sustainable, accessible, AI-powered health monitoring tools that empower both clinicians and patients to track brain health remotely and efficiently. In recent years, large language models (LLMs) like GPT-4 and beyond have revolutionized human-computer interaction. Through intuitive prompt-based systems, these models eliminate the steep learning curves traditionally associated with AI, making sophisticated technology more inclusive—even for those with limited technical expertise. This democratization is especially valuable in healthcare, where clinicians, caregivers, and researchers can leverage LLMs for documentation, decision support, translation of medical jargon, and personalized patient communication. However, true innovation stems not only from technology itself, but from how effectively interdisciplinary teams integrate it. Our diverse team—from neuroscience to software engineering—is acutely aware of the challenges in translating across disciplines. The beauty of prompt engineering is that it acts as a universal bridge, enabling clearer collaboration and lowering the threshold for contributing impactful solutions. Our current efforts are focused on developing intuitive, low-code AI tools that can:

• Assist in early detection of cognitive decline through natural language prompts.
• Enable secure, cost-effective remote patient assessments.
• Foster greater independence for older adults through proactive monitoring and timely interventions.
• Break down barriers to specialty care by expanding healthcare access beyond urban centers. We believe the future of aging healthcare lies at the intersection of empathy, accessibility, and AI. And with the tools now available, that future is within reach.

What it does

MoCA-BOT is an AI-powered chatbot designed to administer selected sections of the Montreal Cognitive Assessment (MoCA), including tasks related to language, memory and delayed recall, abstraction, naming, attention, and orientation. It evaluates user responses in real time, calculates a cumulative cognitive score based on the standardized MoCA rubric, and offers tailored recommendations. When scores fall below key thresholds, MoCA-BOT suggests relevant medical follow-ups and provides access to curated health resources for further support.

How we built it

We utilized prompt engineering with OpenAI-ChatGPT-3 and a Python interface, as well as the basic structure of existing MoCA.

Challenges we ran into

In the early stages of development, we encountered several challenges while working with the ChatGPT-3 interface. One major hurdle was fine-tuning hyperparameters such as temperature, top-p, frequency penalty, and presence penalty to achieve balanced and reliable output. The inherent non-determinism of the model—producing varied responses to identical prompts—also posed difficulties in ensuring consistency for evaluation and deployment. Additionally, crafting prompts with valid logical structure required careful iteration. We had to be particularly thoughtful in selecting precise and unambiguous language to minimize misinterpretation and enhance the chatbot’s responsiveness. These obstacles highlighted the importance of both experimentation and cross-disciplinary understanding when deploying large language models in sensitive use cases like cognitive health monitoring.

Accomplishments that we're proud of

Following an initial review of existing literature and tools, we believe our approach to leveraging large language models (LLMs) represents a novel contribution to healthcare innovation—particularly in reducing access barriers to essential cognitive assessment resources. By applying LLMs in a thoughtful, user-centered way, we've developed a tool that not only supports remote care delivery but also empowers healthcare professionals with minimal technical backgrounds to integrate AI into their workflow. Our project stands as a compelling proof of concept (PoC) for how interdisciplinary collaboration can drive meaningful technological advancement. As an undergraduate team from diverse academic domains, we've successfully merged our unique skill sets to create a solution that is not only technically sound but socially relevant and scalable.

What we learned

From the creation of MoCA-BOT, our team learned a lot about the dynamic nature of chatgpt-3, and how to use prompt engineering, or use of user-generated input prompts, to solve logic structures and to achieve expected outputs. We also gained first-hand knowledge of the interactive nature of LLMs and their ability to explain healthcare ideas in layman’s terms to the end user; thus, making healthcare more accessible for everyone.

What's next for Moca-bot

While MoCA-BOT is capable of completing many aspects of MoCA, there are some limitations that we believe can be improved in the future. For example, long inputs are difficult due to the token maximum set by OpenAI. We would need to use additional programming or other solutions to bypass this token limitation.

Other steps we plan to take include:

Using ChatGPT-4 or more robust LLMs with larger power Fine-tuning and training on "ideal" chat datasets Incorporating other AI models technologies for the Visuospatial and Executive Tasks Recording the chat logs and extracting data for research purposes Incorporating speech-to-text abilities for user input

Built With

• chatgpt
• openai
• python