M.E.D.U.S.A.

Inspiration

Patients with chronic illnesses often take 5 to 10 different medications daily. While doctors are experts at prescribing individual pills, they lack tools to evaluate complex, multi-drug combinatorial effects. This "polypharmacy blind spot" leads to dangerous, unforeseen side effects. We wanted to build a real-time safety net that catches these interactions before the patient ever swallows a pill.

What it does

Our project is a hybrid AI platform that analyzes complex medication regimens in real-time. Instead of relying on human memory to cross-reference dozens of overlapping drug interactions, it mathematically maps out chemical clashes and outputs a simple, plain-English alert for the prescribing doctor.

How we built it

We used a dual-layer architecture to ensure both accuracy and usability:

The "Brain" (ML Graph Layer): We built a structured graph database (using Python) populated with known drug interactions and enzyme pathways. This layer mathematically calculates exact chemical clashes without guessing.
The "Voice" (Generative AI Layer): We connected an LLM to the graph. When the graph flags an intersection (e.g., Drug A and Drug B both inhibit the same liver enzyme, causing an overdose of Drug C), the LLM translates that raw data into a clear, actionable clinical warning.

Challenges we ran into

The biggest hurdle was preventing AI hallucinations. In healthcare, a hallucinated fact can be fatal. We solved this by strictly sandboxing our LLM. It is not allowed to "guess" medical facts; it is only allowed to translate the hard, deterministic data provided by our Graph database.

Accomplishments that we're proud of

We are incredibly proud of building a "zero-hallucination" architecture. By combining strict graph logic with the natural language capabilities of an LLM, we created a tool that is both highly accurate and extremely easy for a doctor to read on the fly.

What we learned

We learned that pure machine learning or raw LLMs aren't enough for healthcare on their own. The real power comes from combining structured data structures (like graphs) with generative AI to create hybrid systems.

What's next

Our next step is building an API to integrate this system directly into standard Electronic Health Record (EHR) software, so it can automatically scan a patient's chart in the background without the doctor having to manually type in the medication list.