Guim Casadellà

Incoming Data Scientist at McKinsey | Data, ML and Edge Systems

Data Science, Machine Learning and AI

I build practical AI, data, and edge systems that connect modeling with reliable products and operational decisions. Previously, I worked as a Data & ML Engineer at Haddock; I will join McKinsey as a Data Scientist in Madrid in September 2026.

Below are selected projects that best showcase my work across applied machine learning, computer vision, optimization, realtime applications, and edge computing.

---

Featured Projects and Experience

1. HackUPC 2026 - guAIta

• Award: 3rd place in the Qualcomm Challenge.
• Overview: guAIta is an Edge AI incident-response system for wild-boar detection around Collserola. Arduino UNO Q stations run computer vision locally, then send evidence-backed detections and telemetry to a realtime operations dashboard with risk scoring, snapshots, live camera viewing, and automated authority escalation.
• My Contribution: Drove much of the end-to-end application and integration layer across the backend, dashboard, device contracts, telemetry, live-stream path, alert handling, and ElevenLabs escalation workflow within a four-person team.
• Technologies Used: Arduino UNO Q, Edge Impulse FOMO, TypeScript, Fastify, React, MapLibre, SQLite, Socket.IO, ElevenLabs
• Key Highlights: Edge AI, On-device Inference, Realtime Operations, Hardware/Software Integration, Automated Escalation

• GitHub

  Devpost

  Video Demo

2. Datathon FME 2024 - Design Decoder MANGO Challenge

• Award: 1st place in the Mango Challenge.
• Overview: Design Decoder automates the process of logging new garment samples into digital systems. The system extracts design attributes from product images and metadata using FashionCLIP embeddings and dedicated machine-learning classifiers.
• Technologies Used: Python, PyTorch, FashionCLIP, XGBoost, OpenCV, Streamlit, NumPy, Pandas, Scikit-learn
• Key Highlights: Image Embeddings, Attribute Classification, Computer Vision, AI Automation

• GitHub

  Kaggle Competition

3. Personal Edge Lab

• Overview: Personal Edge Lab is a local-first edge platform that combines ESP32 sensing and infrared actuation with a RUBIK Pi 3 control plane. It provides durable telemetry, authenticated mobile control, operational alerts, audited commands, and an owner-only Telegram interface.
• Technologies Used: Python, FastAPI, React, TypeScript, SQLite, ESP32-S3, C++, Telegram Bot API, Nginx, systemd
• Key Highlights: Edge Computing, IoT, Local-first Architecture, Safe Physical Control, Telemetry, Operational Reliability
• GitHub

4. HackUPC 2025 - DataCenter-DeCoder

• Award: 2nd place in the Siemens Energy Challenge.
• Overview: DataCenter-DeCoder is an interactive tool for designing and optimizing data-center configurations. It generates valid layouts under complex infrastructure constraints and combines optimization algorithms with a drag-and-drop interface for configuration and validation.
• Technologies Used: Python, FastAPI, React, TypeScript, Vite, MongoDB
• Key Highlights: Constraint-based Optimization, Interactive Module Placement, Resource Management, Data Visualization
• GitHub

5. Reward Optimization in Semantic Segmentation - CVC Internship

• Institution: Computer Vision Center (CVC)
• Internship Period: December 2023 - July 2024
• Overview: Developed a reward-optimization framework for semantic-segmentation tasks, exploring reinforcement-learning techniques for improving model behavior and evaluation.
• Technologies Used: Python, PyTorch, TensorBoard, OpenCV, OpenMMLab, DeepLabv3
• Key Highlights: Semantic Segmentation, Reward Optimization, Model Evaluation, Research Engineering
• GitHub

6. LoRA for Semantic Segmentation Domain Adaptation - CVC Internship

• Institution: Computer Vision Center (CVC)
• Internship Period: December 2023 - July 2024
• Overview: Explored Low-Rank Adaptation for transferring a transformer-based semantic-segmentation model from synthetic to real-world domains.
• Technologies Used: Python, PyTorch, SegFormer-B0, LoRA, Hugging Face
• Key Highlights: Domain Adaptation, Semantic Segmentation, Transformer Models, Parameter-efficient Fine-tuning

• GitHub

  Hugging Face

7. Bank to Notion Finance OS

• Overview: Bank to Notion Finance OS automates the upload, processing, and categorization of bank transactions from CSV files into a Notion financial dashboard.
• Technologies Used: Python, FastAPI, Vue.js 3, Vite, Tailwind CSS, Notion API
• Key Highlights: Notion Integration, Transaction Automation, CSV Processing, Personal Finance
• GitHub

8. Fashion Sales Prediction - Machine Learning Course Project

• Overview: Explored the viability of predicting fashion-product sales from a multimodal dataset using metadata, images, and temporal features.
• Technologies Used: Python, Pandas, PyTorch, OpenCV, Matplotlib, Scikit-learn, XGBoost, SMOTE
• Key Highlights: Multimodal Features, Time Series, Bootstrapping, Regression, Classification

• GitHub

  Dataset

9. HackUPC 2024 - Inditex TECH Challenge

• Overview: Built a fashion system combining duplicate-image detection, outfit recommendation, and out-of-stock assistance. The project used CLIP embeddings and U-Net-based clothing segmentation.
• Technologies Used: Python, PyTorch, Hugging Face, CLIP, U-Net, Semantic Segmentation
• Key Highlights: Image Embeddings, Recommendation, Semantic Segmentation, Frontend

• Video Demo

  GitHub

  Devpost

10. CriminalMap

• Overview: Developed a crime-mapping application using a variation of the A* algorithm to suggest routes that balance travel time with real-time crime data.
• Technologies Used: Python, C, Overpass API, Selenium, BeautifulSoup, Google Maps API, Matplotlib
• Key Highlights: Algorithm Design, Pathfinding, Geospatial Analysis, Data Collection, Visualization
• GitHub

11. Novartis Datathon 2023

• Overview: Developed a machine-learning model for healthcare sales forecasting, with time-series feature engineering for trends, seasonality, lags, and rolling behavior.
• Technologies Used: Python, Pandas, LightGBM, Matplotlib
• Key Highlights: Time-series Forecasting, Feature Engineering, Data Cleaning, Gradient Boosting
• GitHub

For more projects, check out my GitHub repositories.

---

🛠 Skills & Technologies

• Programming Languages: Python, TypeScript, C++, C, R, Java
• Data and Machine Learning: Pandas, NumPy, Scikit-learn, XGBoost, PyTorch, TensorFlow, Hugging Face
• Computer Vision: OpenCV, CLIP, OpenMMLab, Semantic Segmentation
• Applications and APIs: FastAPI, Fastify, React, Vue, Vite, Socket.IO
• Data Systems: SQL, SQLite, MongoDB, Data Pipelines, Experimentation
• Edge and IoT: Arduino, ESP32, Edge Impulse, Local-first Systems

🌐 Connect With Me And Check My Other Work

I’m always open to discussing data science, applied AI, and new technical projects.

• LinkedIn: guimCC
• GitHub: guimCC
• Hugging Face: guimCC
• Email: guim.casadella@gmail.com