MedGuide — Medication & Research

Explore medication guidance, interactions, and scientific studies.

Comment

MedGuide — AI Medication & Supplement Intelligence

Tagline: Understand what you take, what may interact, and what the research actually says.

Live Demo: https://medguide-orcin.vercel.app/evidence-lens

Source Code: https://github.com/enterlina/medguide

Short Description

MedGuide is an AI-powered web platform that helps people understand medication instructions, check potential interactions between medications, supplements, food, and drinks, and explore what scientific research says.

It combines official FDA, DailyMed, and NIH information with scientific publications from PubMed, using AI-powered retrieval and NVIDIA Nemotron through Nebius Token Factory to present findings in a clear, source-linked format.

Inspiration

We all want to make better decisions about our health. But when it comes to medications, vitamins, and supplements, finding clear, reliable information isn't always easy.

Doctors prescribe medications, but choosing supplements is often left to us. And if you're curious about the science behind them, navigating medical research can feel like a project of its own.

That's why we started MedGuide — an AI-powered web platform that helps people understand medication guidance, explore potential interactions, and see what scientific research says about medications and supplements.

What it does

MedGuide brings together two main features:

1. Taking & Combining

  • Check potential interactions between medications, supplements, food, and drinks.
  • Explore warnings, precautions, and timing recommendations.
  • Read medication instructions in a simple, understandable format.
  • Access original FDA, DailyMed, and NIH sources.

2. What Studies Say

  • Find scientific publications from PubMed, PMC, and ClinicalTrials.gov.
  • Read AI-generated summaries of research findings.
  • Explore study designs, participants, and outcomes.
  • View citations, journal metrics, and original publications.
  • Compare available scientific evidence across studies.

MedGuide is designed to make medication information and scientific research easier to access, understand, and verify. It supports informed decisions without replacing professional medical advice.

How we built it

We combined a user-friendly web interface with medical data processing, hybrid retrieval, and AI-powered research synthesis.

Frontend

  • Next.js 16, React 19, TypeScript.
  • Responsive interface with two connected user workflows.

Backend

  • Python, FastAPI, Pydantic, and SQLite.
  • Integration with DailyMed, openFDA, NIH ODS, and RxNorm.

AI & Research

  • PubMed, PMC, and ClinicalTrials.gov for scientific publications.
  • Hybrid search combining BM25 and multilingual E5 with Reciprocal Rank Fusion (RRF).
  • NVIDIA Nemotron 3 Nano through Nebius Token Factory for source-linked research synthesis.
  • Structured evidence extraction, validation, and human-review workflows.

Our approach connects official medical sources with AI-powered analysis while keeping the original evidence accessible.

Challenges we ran into

Medical accuracy and transparency

Medical information requires careful interpretation. We worked on keeping AI-generated summaries connected to original sources and distinguishing verified information from findings that still require review.

Interaction detection

Checking combinations of medications, supplements, and food is complex. Our safety evaluation revealed gaps in interaction coverage, helping us identify areas that need improvement.

Scientific evidence quality

Finding relevant research is only part of the challenge. Study design, participant numbers, outcomes, and limitations all affect how findings should be interpreted.

Making information understandable

We focused on presenting technical medication instructions and scientific findings in a format that people without medical backgrounds can navigate.

Accomplishments that we're proud of

  • Built and deployed a working MedGuide web platform with two user workflows.
  • Implemented medication, supplement, food, and drink interaction checks.
  • Integrated official medication information and scientific research sources.
  • Developed hybrid retrieval and AI-powered research summaries using NVIDIA Nemotron through Nebius Token Factory.
  • Created structured research views with study findings, citations, and publication details.
  • Built a dataset of 7,865 interaction-rule candidates with review status tracking.
  • Achieved deterministic source-passage matching for 7,864 of 7,865 extracted passages.
  • Evaluated retrieval on 323 NFCorpus queries, achieving nDCG@10 of 0.3461 with RRF.
  • Added an audit trail and a human-review workflow to support evidence verification.

We're especially proud of combining medication guidance, scientific research, and transparent AI processing in one accessible application.

Our evaluations also helped identify limitations that we plan to address before considering the platform suitable for clinical use.

What we learned

Building MedGuide showed us that healthcare AI is not just about generating answers.

Reliable sources, transparent findings, thoughtful evaluation, and clear UX are essential when working with medical information.

We also learned the value of testing retrieval approaches instead of assuming that more complex AI pipelines always produce better results.

What's next for MedGuide

  • Expand coverage: Add more medications, vitamins, supplements, and food interactions.
  • Improve interaction checking: Address gaps identified during safety evaluations.
  • Strengthen research analysis: Improve evidence relevance, research summaries, and study comparisons.
  • Personalized guidance: Introduce medication schedules, timing recommendations, and reminders.
  • Scientific validation: Continue working with expert reviewers to evaluate accuracy and reliability.
  • Continuous updates: Keep medication information and scientific publications up to date.

Our vision is to develop MedGuide into a comprehensive medication and supplement intelligence platform, combining interaction checking, personalized medication guidance, and continuously updated scientific evidence in one accessible experience.

Built with

Next.js · React · TypeScript · Python · FastAPI · Pydantic · SQLite · BM25 · multilingual E5 · RRF · Nebius Token Factory · NVIDIA Nemotron 3 Nano · DailyMed · openFDA · NIH ODS · RxNorm · PubMed · PMC · ClinicalTrials.gov

Built With

Share this project:

Updates