Abhinay Karthik
All work

Case 4 of 4, In development

EntraFlux

An AI safety platform for GLP-1 metabolic therapies: predicting side-effects from wearables, spotting counterfeit drugs, mining safety signals.

Heartbeat traces receding in time: the wearable signal it reads before symptoms show.

Overview

EntraFlux is for people on GLP-1 medications such as Ozempic and Wegovy. It reads wearable biometrics to predict nausea and adherence problems before they happen, and gives adaptive guidance in the moment. It also watches online marketplaces for counterfeit listings and mines adverse-event signals from social media and published trials.

Next.js / React Native / FastAPI / PyTorch / PostgreSQL / Redis / FHIR

Problems, and what I built

  1. 01

    Predicting subjective side-effects from objective biometric signals

    PyTorch sequence models trained on HealthKit and Google Fit time series

  2. 02

    Detecting counterfeit listings in messy, unstructured web data

    A multi-modal counterfeit classifier over images, text and metadata

  3. 03

    Normalising data from very different wearable platforms

    FHIR-compliant ingestion adapters for HealthKit and Google Fit

  4. 04

    Designing clinical-grade features before clinical validation

    A research-mode architecture where features ship behind validation gates

What it does

  • Wearable biometric analysis to forecast GI side-effects
  • Adaptive meal timing and anti-nausea guidance in real time
  • Counterfeit drug detection across web marketplaces
  • Safety signals extracted from social media and clinical trials
  • FHIR-compliant data exchange for trial cohort matching

Result

Metabolic therapies become safer and easier to stay on, by surfacing early warnings that neither patients nor clinicians can see today.