Real-world data science applied to environmental, health, and climate challenges — all code open source on GitHub.
⭐ Featured · Air Quality
PM2.5 Air Pollution Forecasting — LSTM
Time-series LSTM neural network forecasting PM2.5 pollution trends on ~50k+ environmental records. Achieved 15–20% accuracy improvement over statistical baselines. Applicable to real-time public health monitoring.
PythonLSTM
KerasTime Series
Climate · Emissions
Global Emissions Analysis — ML
Machine learning analysis of global emissions datasets, identifying patterns and trends across regions and sectors to support climate policy decision-making.
PythonML
EDAClimate data
⭐ Published · Hackathon Winner
Travel Air IQ — Air Quality Tourist Tool
Won ARPA Puglia Data Hackathon. Built a real-time air quality dashboard helping tourists make environmentally informed decisions. Integrated live ARPA Puglia data with interactive visualisations. Published Springer Nature 2024.
PythonPlotly
Real-time APISpringer 2024
Epidemiology · COVID-19
Human Mobility & COVID-19 Analysis
Analysed ~20k+ population mobility and COVID-19 infection records across Italian regions. Regression and statistical modelling quantified effectiveness of non-pharmaceutical interventions. Contributed to 2 peer-reviewed publications.
PythonRegression
Statistical AnalysisSpringer 2021
Classification · Water
Water Quality Pollution Classification
Classified water pollution levels using Logistic Regression and Random Forest models. Demonstrates ML versatility beyond air quality domain — applicable to environmental monitoring and regulatory compliance.
Pythonscikit-learn
Random ForestClassification
Deep Learning · Telecom
Telecom Churn Prediction
Deep learning model predicting customer churn from telecom usage data using Keras. Demonstrates ML breadth across non-environmental domains — applicable to commercial data analyst roles.
PythonKeras
Deep LearningClassification
⭐ Live Demo · Agentic AI
Wetland Carbon Monitoring Agent — Copilot Studio
A live AI-powered environmental monitoring agent that analyses wetland ecosystem health, carbon flux patterns, and sink-to-source transition risk. Built using Microsoft Copilot Studio on methodology from EU-funded LifeWatch Italy research and FluxNet open data.
Agentic AI
Microsoft Copilot Studio
FluxNet data
LifeWatch Italy
Live Demo
⭐ Multi-Agent System · Python · Groq LLM
Wetland Carbon Monitoring — Multi-Agent System
A fully coded multi-agent AI system with 4 specialist subagents: Risk Assessment Agent, Data Fetching Agent (live OpenAQ), Report Generation Agent, and Alert & Intervention Agent — all coordinated by an Orchestrator. Built using Groq LLM (Llama 3.3 70B), Streamlit, and Python. Grounded in EU LifeWatch Italy PhD research. Users bring their own API key.
Python
Groq LLM
Multi-Agent
Streamlit
OpenAQ API
Subagents
LifeWatch Italy