Skip to content

nihithasrininivasan/PharmaGaurd

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

29 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🧬 PharmaGuard — Clinical AI for Pharmacogenomic Risk Intelligence

PharmaGuard is an end-to-end pharmacogenomics decision-support platform built during a high-pressure hackathon to transform raw genomic data into fast, explainable, clinician-ready insights.

While many teams focused only on AI generation or UI dashboards, we engineered a clinically grounded system — combining genomics parsing, CPIC-aligned risk modeling, and safety-focused AI explanations into a single real-time workflow.


🚀 What PharmaGuard Does

Upload a patient VCF file + select a drug → PharmaGuard instantly delivers:

  • 🧬 Gene + Diplotype interpretation
  • ⚠️ CPIC-aligned risk classification
  • 📊 Pharmacogenomic profile with detected variants
  • 🤖 Grounded AI clinical explanation (mechanism-aware)
  • 🔥 Risk Heatmap visualization
  • 💬 “Ask PharmaGuard” clinician assistant

All powered by a privacy-first local LLM pipeline.


🏆 Why PharmaGuard Stands Out (Even Among Similar PS Builds)

Many hackathon solutions stopped at “AI generates medical text.” PharmaGuard goes further — focusing on clinical reliability, explainability, and real workflow design.

✅ 1. Clinically Grounded AI — Not Generic LLM Output

  • AI responses are constrained by structured pharmacogenomic data
  • Variant citations and biological mechanisms included
  • CPIC-aligned reasoning enforced
  • Hallucination-resistant prompt architecture

⚡ 2. Real-Time Non-Blocking Architecture

  • Risk engine responds instantly
  • LLM explanation loads asynchronously
  • Ultra-Turbo async pipeline reduces perceived latency
  • Model warm-keep and token-limited inference for speed

🔥 3. Risk Heatmap Mode (Hackathon Differentiator)

Dynamic UI highlighting toxicity intensity using genomic risk signals — translating complex PGx data into an immediately interpretable clinical visual.

💬 4. Ask PharmaGuard — Context-Aware Clinical AI

Unlike basic chatbots:

  • Grounded in current patient genotype + drug
  • Mechanism-focused answers
  • Clinician-tone generation
  • Safety disclaimers enforced

🛡️ 5. Safety-First Design Philosophy

  • Structured schema validation
  • Fallback explanation logic
  • No hallucinated genes or variants allowed
  • Built for decision support, not autonomous prescribing

🧠 Tech Stack

Backend

  • FastAPI
  • Pydantic
  • Ollama (Llama3 local model)
  • Async pipeline architecture

Frontend

  • React + Vite
  • Real-time state updates
  • Clinical heatmap UI system

AI Layer

  • Local LLM inference
  • Structured prompt grounding
  • Non-blocking async explanation pipeline

⚙️ Architecture Highlights

VCF Upload → Variant Parser → Risk Engine → Clinical Recommendation
                              ↓
                        Async LLM Explanation
                              ↓
                    Heatmap + Clinician UI

🧪 Key Engineering Achievements During Hackathon

  • Built full pharmacogenomic pipeline from scratch
  • Integrated local LLM safely into clinical workflow
  • Designed hallucination-resistant explanation system
  • Implemented real-time UX despite heavy AI inference
  • Delivered complete frontend + backend integration under time constraints

🎥 Demo

👉 Full walkthrough video included with this submission.


🙌 Acknowledgements

Massive thanks to RIFT by PWIOI for hosting an incredible hackathon environment that encouraged deep technical execution, clinical responsibility, and bold experimentation.


⚠️ Disclaimer

PharmaGuard provides AI-assisted pharmacogenomic insights for educational and clinical support purposes only. Final prescribing decisions must be made by licensed healthcare professionals.


💥 Vision

PharmaGuard is not just a hackathon project — it’s a step toward making pharmacogenomics interpretable, actionable, and safe in real clinical workflows.

Because precision medicine shouldn’t feel like decoding a genome alone.

About

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors