Hey! I'm David 👽
I’m a Data Scientist who loves building and implementing innovative and highly technical data projects in order to change the world.
I love using data to read between the lines and find possibilities not detectable from simple analysis. As per my double degree education, I am also a Computer Scientist and Engineer, allowing my data skills, decision-making and problem-solving to go beyond average and find new approaches.
- Currently studying a Double Bachelor:
Data Science - TU Eindhoven and Tilburg University
Computer Science & Engineering - Tilburg University
- You can reach me at: david.manda.loureiro@gmail.com
These are some of the tools I have worked most with for professional work, university in depth data challenges and personal projects..
Other DS / ML tools:
Seaborn · plotext · Optuna · NLTK · spaCy · VADER · GeoPandas · Folium · Leaflet · classical ML methods, feature engineering, evaluation & experimentation
Topics & frameworks:
Computer vision (semantic segmentation, U-Net, EfficientNet encoders) · classical ML pipelines · hyperparameter optimization (Optuna) · interpretability & feature-based modeling
Other DE / MLOps:
Dask · data pipelines · model deployment basics · experiment tracking & reproducible workflows
Other tools:
Plotly Dash · Streamlit · custom Flask dashboards · interactive geospatial dashboards (GeoPandas, Folium, Leaflet)
Concepts:
RESTful API design · backend-for-frontends for dashboards · integration of ML models into APIs & web apps
Topics:
Data Structures & Algorithms · concurrency · computer systems · networking basics · packet sniffing & analysis (Wireshark)
- PlusCal & TLA+ (algorithm specification and verification)
- Simulation of autonomous systems / earthquake-survival search robots (simulation platform to be filled in)
- NLP pipelines: NLTK · spaCy · VADER · translation via DeepL API
These are the most interesting projects I have worked on that show the value I can provide and the insightful and creative thinking I am capable of, please feel free to go more into detail if you find any of them interesting.
This project was an investigation to aid Reef Support, a company focused on addressing coral health around the globe and find new or improved uses for machine learning to reduce labour-intensive tasks and increase bleaching detection accuracy.
🔍 Details
What it is
Capstone Data Challenge project (TU/e JBG060) in collaboration with Reef Support. The goal is to move beyond binary “bleached / not-bleached” classification and instead build a low-compute, explainable pipeline that segments coral in photo-quadrats and derives a graded health signal from interpretable image features. :contentReference[oaicite:0]{index=0}
We use a U-Net with an EfficientNet-B0 encoder to obtain coral / non-coral masks (mIoU ≈ 0.67 on held-out reef sites). Within these masks we compute texture and color/whiteness features (e.g. Laplacian variance, LBP, GLCM correlation, albedo, luminance, saturation, red channel) and aggregate them into (1) a linear health index and (2) percentile-matched exemplar images to support human review and decision-making around bleaching.
Tech stack
- Data & modeling: Python (Jupyter), U-Net + EfficientNet-B0 segmentation, linear regression for health index
- Feature engineering: Classical CV & texture descriptors (Laplacian, LBP, GLCM, HSV/RGB features)
- Experimentation: Site-stratified evaluation, Optuna for hyperparameter search, mask QC & preprocessing :contentReference[oaicite:2]{index=2}
Key features
- Coral vs. background segmentation with U-Net (EfficientNet-B0 backbone), achieving ~0.67 mIoU on held-out reef sites
- Interpretable feature extractor (3 texture + 6 color/whiteness features) computed only inside coral masks to track paleness and micro-texture changes related to bleaching :contentReference[oaicite:3]{index=3}
- Two complementary health outputs: a transparent regression-based health index (0–100) and percentile-matched exemplar retrieval to visually compare similar health states and aid explainability
What I learned
-
🟢 How to design and evaluate an end-to-end segmentation + feature-engineering pipeline under domain shift (site-stratified splits, normalization, mask quality checks, and correlation-based feature pruning).
-
🟢 How to make ML models more responsible and trustworthy in an environmental context: handling dataset bias (standardization, robust statistics), keeping models explainable, and thinking about how responsibility is shared between developers, agencies, and local communities when AI informs high-impact decisions. :contentReference[oaicite:6]{index=6}
Links
- Repo: github.com/ML-Coral-health-estimation
- Report: github.com/ML-Coral-health-estimation/Capstone Report
- Legal/Social/Ethical Essay: github.com/ML-Coral-health-estimation/Capstone Essay
- Poster: https://github.com/DavidMandado/DavidMandado/blob/main/JBG060%20Poster%20(2).pdf
- Reef Support: https://www.reef.support/
Images of the data results:
;
Short one-liner here.
🔍 Details
What it is
[2–3 sentence explanation]
Tech stack
- [Tech]
- [Tech]
Highlights
- 🚀 Highlight 1
- 🚀 Highlight 2
Links
- 🔗 Repo: https://github.com/YOUR_USERNAME/PROJECT_2_REPO
- 🌐 Live demo: https://PROJECT_2_LIVE_URL
Short one-liner.
🔍 Details
Problem it solves
- [Explain the problem]
My approach
- [Explain how you solved it]
Links
⚡ GitHub Stats
🔥 GitHub Streak
🧩 My Setup
- OS: [e.g. macOS / Windows / Linux]
- Laptop: [your machine]
- Browser: [your browsers]
- Terminal: [e.g. Zsh, PowerShell, etc.]
- Editor: [e.g. VSCode, JetBrains etc.]
- Other Tools: [Postman, Notion, Figma, etc.]
- To stay updated: [e.g. Twitter, Reddit, Hacker News]
If you find my work useful, consider:
- ⭐ Starring some of my repositories
- 💬 Reaching out for collaboration or opportunities
Thanks for stopping by! 😊