End-to-End Microsoft Fabric solution with Git Documentation
Skills: PySpark, SQL, DAX, Power BI, Medallion Architecture, ETL.
Business Impact: Identified 81,744 MWh in recorded efficiency risk, translating technical waste into €4,904,640 of avoidable costs.
Key Achievement: Engineered a real-time "Single Source of Truth" that maintains 102.59% heat self-sufficiency while tracking ESG targets across a diverse energy fleet.
By leveraging DirectLake, I bypassed the traditional 'Import' vs 'DirectQuery' trade-off, allowing executive users to see massive energy load changes instantly without the overhead of scheduled refreshes.
BIP HMI Dashboard - Anomally detection
BIP HMI Dashboard - AI Activated
Transforming Telemetry into Autonomous Intelligence
In high-stakes biomanufacturing, telemetry data is often siloed in static logs, leading to a reactive operational posture where deviations are discovered only post-batch. This gap creates significant risk for high-value pharmaceutical yields.
I developed the Bioprocess Insight Platform (BIP): a containerized, AI-augmented Triad Microservice Architecture designed to serve as a high-fidelity Digital Twin for Sartorius Biostat® fermentation systems. Unlike a simple dashboard, BIP orchestrates three independent Dockerized layers to provide a closed-loop, compliant solution:
Autonomous SCADA Simulation (AI Pilot): A high-frequency FastAPI engine that streams multivariate sensor data and features an integrated AI Pilot. This layer monitors batch health in real-time and automatically triggers corrective Impeller RPM adjustments to restore Dissolved Oxygen levels via physics-based coupling.
GxP Compliance & Governance: A dedicated .NET 8 microservice that ensures 21 CFR Part 11 data integrity. It features Role-Based E-Signatures (Lead Scientist, Lab Tech, QA) and unique Batch ID tracking (B2026-XXX) to ensure every production run is isolated and attributed.
Predictive HMI: A reactive Node 22/TypeScript dashboard that translates raw telemetry into predictive Batch Health Scores. It utilizes moving-window linear regression to project temperature trends 60 seconds into the future, enabling proactive intervention.
By architecting this polyglot solution, I have demonstrated a production-ready approach to transforming raw process data into a proactive, observable, and compliant digital twin. All code is version-controlled and fully documented.
Backends (Polyglot Layer):
Data Engine: Python 3.12, FastAPI, Pandas (High-frequency telemetry & AI Pilot logic)
Compliance/Audit: .NET 8, C#, ASP.NET Core (Immutable GxP logging)
Frontend (Industrial HMI):
Stack: React 18 (TS), Node 22, Recharts, Lucide-React (Real-time viz & iconography)
Styling: CSS-in-JS, Custom Keyframes (Deterministic GxP pulsing alarm animations)
Data & Persistence:
Storage: SQLite (Local/Edge), EF Core (Strict GxP schema enforcement)
Optimization: Encapsulated CSV datasets for high-speed containerized data locality
Infrastructure & DevOps:
Orchestration: Docker & Docker Compose (Multi-container parity & service isolation)
CI/CD: GitHub Actions (Automated build/deploy pipelines)
Cloud: Vercel (Frontend), Render (Distributed Backends)