All projects
Artificial Intelligence · Final-Year ProjectIn progress

Pharmaceutical Supply-Chain Digital Twin

PharmaTwin · SLIIT final-year research project · J26-IT-353

An AI-driven digital-twin framework for resilient pharmaceutical supply-chain optimization, integrating multi-source data, market intelligence, predictive analytics and intelligent warehouse management.

View on GitHub(opens in a new tab)

This is my ongoing final-year research project. The framework is still being designed and developed, so the content below describes its goals and proposed approach rather than finished results.

Overview

PharmaTwin is a research project that proposes an AI-driven digital twin of the pharmaceutical supply chain — a connected, data-driven model that aims to help the supply chain anticipate disruptions, plan demand and manage expiry-sensitive stock more intelligently.

  1. 01Digital-twin architecture and simulation
  2. 02Multi-source data integration and market intelligence
  3. 03Demand forecasting and predictive analytics
  4. 04Intelligent warehouse FEFO and QR spatial management

Problem statement

Pharmaceutical supply chains have to balance strict shelf-life and storage requirements against uncertain demand and external disruptions. When signals about global events, stock levels, expiry dates and demand sit in separate places, it becomes difficult to anticipate shortages and act early.

Objectives

  • Design a digital-twin architecture that represents the state of the supply chain and supports simulation.
  • Integrate multi-source data, including market and event intelligence.
  • Explore demand forecasting and predictive analytics to support planning decisions.
  • Support warehouse operations through First-Expiry-First-Out (FEFO) dispatch and QR-based spatial management.

Technologies

  • Python
  • Machine Learning
  • NLP
  • FastAPI
  • Next.js
  • Node.js
  • Supabase

Planned data sources

  • Global health and disruption feeds (for example WHO, ReliefWeb and GDACS)
  • News and market-intelligence sources
  • Inventory, batch and expiry records
  • Historical demand and order data

Approach

  1. 01

    Architecture

    Design the digital-twin architecture that links data, models and decision dashboards.

  2. 02

    Integration

    Build pipelines that collect and normalise data from multiple external and internal sources.

  3. 03

    Intelligence

    Apply NLP and predictive models to surface disruption signals and forecast demand.

  4. 04

    Operations

    Translate insights into warehouse actions such as FEFO dispatch and QR-based location management.

Proposed architecture

Subject to change as the research progresses.

  1. Sources
    • Event & news feeds
    • Market signals
    • Inventory & orders
  2. Ingestion & API
    • Node.js / Express
    • Validation
    • Schedulers
  3. AI services
    • FastAPI
    • Classification
    • Forecasting
  4. Data layer
    • Supabase
    • PostgreSQL
  5. Interface
    • Next.js dashboards

Results

Results will be shared here once the framework has been evaluated.

Key research challenges

  • Combining heterogeneous data sources that differ in format, reliability and update frequency.
  • Keeping forecasts and recommendations explainable enough to support real operational decisions.
  • Representing expiry-sensitive inventory accurately within a digital twin.
Next projectMaterial Availability DashboardData Analytics · Business Intelligence