MediGuide AI

• 

  the user ask about a clinical question, the identity is confirmed first to give role based information

• 

  since user is a member of the general public, information given is simple to understand. The assistant gives educational information only

• 

  the user asks a question inline with self-medication, the assistant refuses and advices seeking consultation from a doctor

• 

  the user attempts to switch roles, since the role is locked for the whole session once stated, the assistant refuses role switching

• 

  here user asks a clinical question as well, assistant answers without suggesting medications or giving self diagnosis

• 

  here user asks a question in Chinese and the assistant replies in the same language as set in system

Inspiration

My motivation stems from a personal encounter. After developing a skin rash, I uploaded a photo of it to an AI platform, which recommended calamine lotion and chlorpheniramine tablets. Trusting this advice, I proceeded with the suggested treatment, but my symptoms worsened. I later discovered, through a visit to a dermatologist, that the rash was actually a severe fungal infection requiring different care. This experience opened my eyes to the dangers of relying on AI for medical guidance without professional oversight. It underscored the importance of systems that verify user identity, restrict advice within professional boundaries, and refer users to licensed healthcare providers when needed. This insight directly influenced MediGuide AI’s mission to prevent self-diagnosis and unsafe self-medication, which can lead to drug interactions, dietary complications, and even organ damage,

What it does

The system leverages Google’s Gemini AI to analyze user intent, distinguishing whether a query relates to lifestyle advice, medical conditions, medication information, clinical decision-making, or educational purposes. Depending on both this intent classification and the user’s stated role, the application intelligently tailors the depth, complexity, and safety limits of its responses.

For members of the general public, the system delivers broad, non-diagnostic health information and actively blocks advice that could enable self-medication or the misuse of clinical knowledge. Student users receive academically oriented explanations that foster structured learning, but the system avoids providing guidance on treatment decisions. Licensed professionals, meanwhile, gain access to detailed, reference-level discussions, though the system withholds patient-specific recommendations.

Additionally, the application accommodates multiple languages, ensuring that safety protocols remain robust and consistent regardless of the language used. Through strict enforcement of role-based permissions and intent-based response filtering, the system reduces the risk of medical misinformation and supports the responsible use of AI in healthcare education.

How we built it

I built this using Google AI Studio, a platform that allowed me to work directly with a ‘digital brain’ to create a specialized tool—without writing any manual code. I essentially ‘taught’ the AI how to serve as a health assistant.

The secret to the app lies in what’s called System Instructions. Think of these as the AI’s ‘internal manual.’ I programmed it with strict rules that require the AI to first identify who is speaking (a student, a doctor, or a patient), and then keep that person within a specific ‘safety zone’ for the duration of the chat.

I designed three different versions of the same information:

• For the public, I built ‘safety filters’ to ensure the AI never gives dangerous advice. • For students, I instructed it to act like a ‘professor’ and explain the science behind the information. • For professionals, I told it to act like a ‘peer’ and focus on technical guidelines.

I set up the app so that even if a regular person tries to get ‘doctor-level’ information, the AI’s internal rules will prevent it.

Since the AI I used is multimodal, I was able to build a feature where a student can upload a diagram from a textbook and the AI will explain it. However, I specifically blocked that same feature for the public, so they can’t use it to try to diagnose themselves.

Challenges we ran into

I encountered several difficulties while creating my system instructions. Each time I thought I was finished, I would realize there was something else that needed to be corrected. As a result, my prompt came together in bits and pieces, which was not ideal given the approaching submission deadline. I also struggled to rephrase some of my instructions clearly, especially for non-clinical topics or situations where users are denied access to certain information—developing effective system instructions in these cases was particularly challenging.

Additionally, I needed to include disclaimers based on user identity, the type of question asked, and the response provided, which was very difficult to frame appropriately. One major challenge was preventing users from bypassing safety rules by changing roles or rephrasing their questions. Another significant difficulty was finding the right balance between providing educational depth for students and professionals and preventing unsafe medical decisions for the general public. Navigating the usage limits of free AI tools proved particularly challenging during development. To work within these constraints, I had to optimize prompts, conduct efficient testing, and plan API usage strategically. This experience gave me valuable insight into the real-world limitations developers encounter when building scalable AI solutions.

Accomplishments that we're proud of

As a student in the healthcare sector, I am proud to have translated a real public health concern into a practical AI solution despite significant constraints. This project reflects my commitment to ethical AI use in healthcare, patient safety, and responsible innovation. It also marks a meaningful step in my development as a healthcare student dedicated to making a positive impact on society in the field of health.

What we learned

I learned how to design AI systems responsibly, especially in sensitive domains like healthcare. This project improved my understanding of prompt design and how to manage AI limitations. Dealing with platform restrictions and technical setbacks also strengthened my adaptability and problem-solving skills as a student.

What's next for MediGuide AI

Moving forward, MediGuide AI will focus on implementing secure identity verification to reinforce role-based access. Possible approaches include verifying healthcare professionals' credentials, linking user verification to institutions, and implementing layered access permissions. These steps aim to enhance trust, privacy, and overall safety within the platform.

Built With

• google-gemini-ai
• google-studio-ai