Feature: Add safety shield to detect drug-drug interactions when adding medications

[VaidehiSakalkar]

Problem

MedSync-AI currently allows users to add medications without checking for drug–drug interactions, which is risky for users on multiple medications

Proposal

Introduce a lightweight “Safety Shield” that checks a new medication against the user’s existing regimen and warns (or blocks) if a clinically significant interaction is detected. It would automatically warn users if a new medication conflicts with their existing ones.

Approach

• Check interactions between the new drug and existing medications
• If conflict exists:
• Return interaction details + severity
• Allow user to review and optionally override
• Cache interaction results to avoid rate-limit issues

Why this fits MedSync-AI

• Improves user safety without being intrusive
• Differentiates MedSync-AI from basic reminder apps
• Aligns well with the project’s medical-AI direction

If approved, I’d love to convert this into an issue and work on it as SWoC'26 contributor .

[tirth-patel06]

@VaidehiSakalkar
This is a good idea, but for this, it includes the Valid drug data and needs a specific ML model for it or an LLM call to check it.

Are you sure you are fully implementing this, as this is quite hard to implement
If you're sure, just tell me your basic approach, and I will open an issue for it.

[VaidehiSakalkar]

This feature does not rely on ML or an LLM. The approach is fully rule-based and uses an authoritative medical dataset(public medical database (RxNav))

Basic approach:

• When a user adds a new medication, the backend converts the medicine name into a standard medical ID as followed by RxNav
• We then query RxNav’s drug–drug interaction endpoint to check whether the new drug interacts with any existing medications the user is taking.
• If an interaction exists, we can return clinically defined severity (e.g. major / moderate / minor), which we forward to the UI as a warning.
• The user can then cancel or explicitly proceed.

[tirth-patel06]

Thanks for explaining the approach 🙂

This is a solid feature, and the method looks reliable. I’ll open a new issue based on this discussion.
Please add a comment on the issue so it’s easy for me to assign it to you.

[tirth-patel06]

@VaidehiSakalkar
Issue #71 open