Hello, I'm
Shpetim Gashi
Software Developer & Data Scientist
I deliver end-to-end solutions combining web development and machine learning: database-backed applications (Python backends, React/Next.js frontends) and ML systems (forecasting, reinforcement learning, optimisation). I have led an applied AI research project from requirements and method selection through experimentation and evaluation. Currently completing an M.Sc. in Data Science and Intelligent Analytics.
Featured Projects
Reinforcement Learning for Building-Control Research
Simulation-based reinforcement-learning experiments for control problems in building technology. PPO agents trained with stable-baselines3 and Gymnasium, Radiance-based simulation tooling for the environment, and TensorBoard for experiment tracking - with simulation assumptions and limitations kept explicit.
Manufacturing Optimisation & Scenario Planning
Decision-support tool that formulates a constrained manufacturing problem as a linear program in PuLP and exposes scenario analysis through Streamlit and Flask interfaces. Packaged with PyInstaller to run without a Python installation, with the optimisation logic covered by pytest.
Structured Comfort Feedback Platform
A university software-engineering project built around a real-world building-technology use case: a maintainable Flask application for collecting and managing structured user feedback, covering authentication and role handling, PostgreSQL persistence, administrative workflows, REST integration, automated testing, containerisation and CI/CD.
Product Configuration & Augmented-Reality Platform
Customer-facing platform for configuring products and visualising them in augmented reality. Contributed across UI/UX and user flows, authentication, database-backed features, CI/CD and quality review, in a TypeScript/Node stack. Agentic coding workflows were used with human review before integration.
Demand Forecasting & Model Benchmarking
Time-series forecasting pipeline built in two passes: ARIMA/SARIMA baselines with statsmodels first, LSTM models in TensorFlow/Keras about a year later. Evaluation was baseline-oriented throughout, asking whether the added model complexity was justified at all.