Lewis NDAMBIRI | Industrial Engineer & AI/Data Science Specialist
Dual Master's in Industrial Engineering & Computer Science | Optimization, Data, Automation
My Portfolio
Hi, I’m Lewis NDAMBIRI 👋
I bridge systems optimization with scalable software to solve real-world inefficiencies.
🔍 I focus on:
- Process optimization & simulation
- Data-driven decision systems
- Automation & algorithmic efficiency
- Human-centered system design
📧 Email me | 🔗 LinkedIn | 💻 GitHub | 📥 View My Resume
🚀 Featured Projects
1. Maintenance Mode Prediction
Statistics & Data Analysis | Python, Scikit-learn, Pandas
Built a Decision Tree classifier to predict industrial machine operational states (Failure vs. Production) using sensor and maintenance data. Achieved 96% accuracy after hyperparameter tuning.
Key predictors: Sensor5, Age, and Sensor2.
2. Demand Forecasting for Retail Stores
Time Series Forecasting | Python, XGBoost, Pandas, Scikit‑learn, Streamlit, Docker
End‑to‑end forecasting pipeline using XGBoost with leakage‑conscious feature engineering (lag, rolling means, calendar, business features), time‑based holdout validation, and a recursive forecast loop for future unseen dates. Built on the Rossmann Store Sales dataset (1,115 stores, 1.0M+ rows). Achieved 9.23% MAPE, 0.914 R², and a 55% RMSE reduction compared to the best naive baseline. Includes a Streamlit dashboard, Docker support, unit tests, and CI. The model chooses between XGBoost and scikit‑learn fallback automatically.
3. UrbanPulse – Event & Logistics Planner
Service-Oriented Architecture (SOA) | Python, Flask, PostgreSQL, Docker, REST, Telegram Bot
Built a process-centric, 4-layer SOA system that enables users to discover events and receive weather-aware transport recommendations via a Telegram chatbot. Implemented two orchestration services to manage event discovery and commitment workflows, integrating real-world APIs for events, weather, and routing.
- Architecture: Strict 4-layer SOA with independent microservices
- Integrations: Ticketmaster, OpenWeatherMap, OpenRouteService
- Highlights: Preference-based ranking, transport optimization, fully containerized system
It provides a seamless, end-to-end event planning experience from discovery to logistics for users in real-time.
4. Fedrigoni Industrial AI Challenge - Team 3
Industrial AI for Predictive Manufacturing | Python, Pandas, NumPy, XGBoost, Matplotlib, Seaborn
- Project Overview: This project is part of University of Trento’s AI and Innovation course where we focused on improving the slitting process in paper manufacturing at Fedrigoni’s Arco plant. By leveraging predictive AI models and data analytics, we aimed to predict daily slitting output and optimize production management to improve operational efficiency and customer satisfaction.
- Technologies Used: Python, Pandas, NumPy, XGBoost, Matplotlib, Seaborn
- Key Results: Achieved 95% accuracy in forecasting daily slitting output, improved operational efficiency, and helped shift from reactive to predictive production management.
- AI Challenge Provider: The project was part of the Industrial AI Challenge, a platform designed to foster innovation in industrial AI applications. Learn more about the challenge and its initiatives.
→ Case Study → Certificate of Attendance
5. End-to-End Retail Analytics Engine
PostgreSQL, Python ETL, Advanced SQL, Power BI
Designed a complete analytics pipeline - from synthetic data generation in Python to a star-schema data model in PostgreSQL, advanced SQL analysis (cohort retention, RFM segmentation, window functions), and an interactive Power BI dashboard. Answers key business questions: Which products drive profit? How do customers behave? Are we hitting targets?
6. Parallel Numerical Integration using Romberg’s Method
High-Performance Computing (HPC) | C, MPI, OPENMP, PBS, Linux HPC Cluster
Parallelized Romberg integration - a high-accuracy numerical method - using MPI, OpenMP, and hybrid MPI+OpenMP on a 6,092-core cluster using a computationally heavy integrand (sin(x)·e⁻ˣ²) with 1,000,000 artificial iterations to emulate real-world HPC workloads. Evaluated strong/weak scaling, PBS placement strategies (pack, scatter, :excl) and:
- Achieved 31.77× speedup on 32 cores with 99.3% efficiency (near-perfect scaling!)
- Analyzed 50+ benchmark configurations across multiple problem sizes
- Validated theoretical limits using Amdahl’s Law (4.8% serial fraction)
7. MeetScheduler – AI‑Assisted Meeting Scheduler
Full Stack, AI Challenge | React, Node.js, Express, Google Calendar API, Nodemailer, CSS
Built for HackaPrompt AI 2026 at the University of Trento – a Doodle‑inspired meeting scheduler developed entirely with AI assistance. Users create time‑slot polls, vote on availability, and automatically detect the best slot via a quorum. Integrates Google Calendar to show the creator’s busy times and sends email invitations with Nodemailer.
The project doubles as a critical examination of AI programming: where LLMs excel, where they fall short, and what still requires human judgement.
📚 Education
M.S. in Industrial Engineering
Centrale Nantes, 2026
Focus: Operations Research, Enterprise Modeling, Discrete Event Simulation, Product Management, Innovation Engineering, Finance & Economics for Engineering, Project Management.
M.S. in Computer Science
Università degli studi di Trento, 2026
Focus: HPC for Data Science, AI & Innovation, Service Design & Engineering, Innovation & Entrepreneurship studies in ICT.
© 2025 Lewis. Built with ❤️ and GitHub Pages.