Publication
Adversarial Machine Learning: How to Attack and Defend ML Models
https://www.toptal.com/developers/machine-learning/adversarial-machine-learning-tutorial
Pau is currently unavailable
Pau Labarta Bajo
Verified Expert in Engineering
Mathematical Modeling Developer
Barcelona, Spain
Toptal member since April 24, 2019
Bio
Pau is a data scientist and ML engineer with over eight years of experience. He has a passion for building ML-based solutions, from development to deployment. He loves transforming an idea into a model and a model into an API or product. Pau has worked on different problems: financial derivative pricing, digital marketing analytics, deep learning for art generation, or demand prediction for online shopping. His background is in pure mathematics, and he has strong coding skills in Python.
Portfolio
EasyHealth
Google Cloud Platform (GCP), Python
Cogsy Limited
Amazon Web Services (AWS), Python 3, DeepAR
Speakeasy Labs
REST APIs, Segment
Experience
- Mathematical Modeling - 11 years
- Quantitative Finance - 8 years
- Time Series Analysis - 8 years
- Python - 7 years
- Natural Language Processing (NLP) - 3 years
- Large Language Models (LLMs) - 3 years
- Real-time Data - 3 years
- Amazon Web Services (AWS) - 2 years
Preferred Environment
Amazon Web Services (AWS), PyCharm, MacOS
The most amazing...
...model I've built is a generative neural network that creates realistic profile pictures for football players in a mobile game.
Work Experience
Machine Learning Engineer
2021 - PRESENT
EasyHealth
- Developed ML-based bidding bot to acquire leads more cost-effectively.
- Developed a churn prediction model to anticipate policy churn and increase customer retention.
- Built a simulation engine to optimize key parameters for daily operations.
Technologies: Google Cloud Platform (GCP), Python
Time Series ML Engineer
2021 - 2021
Cogsy Limited
- Validated and improved the forecasting methodology that powers Cogsy's app.
- Built an in-house Python package for fast experimentation, leveraging Amazon Forecast AutoML, and custom feature engineering.
- Developed ad-hoc predictive models for several of Cogsy's clients.
Technologies: Amazon Web Services (AWS), Python 3, DeepAR
Data Engineer
2020 - 2021
Speakeasy Labs
- Increased the robustness of the marketing analytics pipeline.
- Helped define and implement an event tracking system adapted to the new iOS 14 tracking restrictions.
- Advised the client on specific low-level details related to Segment.io.
Technologies: REST APIs, Segment
Machine Learning Engineer
2020 - 2020
Lola Market - Freelance
- Developed, deployed, and maintained an ML model to improve the efficiency of the shoppers' fleet.
- Bootstrapped the company's first data warehouse and reporting layer, including Amazon Redshift, Amazon Database Migration Service (DMS), and Tableau.
- Developed several dashboards to help the client improve its fleet management efficiency.
Technologies: Amazon Web Services (AWS), Python, Scikit-learn
Machine Learning Engineer | Statistician
2020 - 2020
Toptal Client
- Analyzed financial market valuations in the Gulf region using explainable machine learning.
- Wrote a Python package to ensure the in-house reproducibility of each step of the analysis, including data processing, data validation, data visualization, model construction, model validation, and model explanation.
- Benchmarked a range of ML solutions and fine-tuned them to enhance model accuracy and explainability.
Technologies: Shapely, Statistics, Scikit-learn, Machine Learning, Python
Explainable AI Engineer
2020 - 2020
15kay
- Supported the development of a scientific Python package in the medical field.
- Researched applicability of the package inside the open-source ML and AI ecosystem.
- Created tutorial notebooks to showcase potential uses of the package.
Technologies: Explainable Artificial Intelligence (XAI), TensorFlow, Jupyter, Python
Data Scientist | Data Engineer
2019 - 2020
Goguru Consulting
- Deployed the client's first data warehouse and data reporting system.
- Developed components of the analytics stack from scratch using Python, SQL, AWS Redshift, and Tableau Online.
- Developed a machine learning model to increase the operational efficiency of Lola Market, a client of Goguru. Lola Market offers its customers the possibility to buy groceries online and have them delivered to their homes in a matter of hours.
Technologies: Random Forests, Scikit-learn, Amazon Web Services (AWS), Tableau, Python, AWS Database Migration Service (DMS), Redshift
Data Visualization | Data Engineer
2019 - 2020
Cyngn
- Created, updated, and maintained the front-end dashboards of the data analytics stack at Cyngn.
- Developed quick visualization prototypes in Tableau and deployed them into dashboards accessible to the engineering team.
- Developed components of the internal ETL tool in Python and SQL.
- Helped back-end engineers integrate front end and back end of the stack inside Amazon Redshift.
Technologies: Amazon Web Services (AWS), SQL, Tableau, Redshift
Mathematical C++ Developer (Genetics Project)
2019 - 2019
Confidential
- Reviewed and documented the proprietary algorithm that performs base calling.
- Advised the client on how to improve the current algorithm.
- Debugged the code and proposed improvements to increase accuracy.
Technologies: C++, OpenCV
Machine Learning Engineer
2019 - 2019
Toptal Client
- Developed statistical and machine learning models to understand the market valuation of financial institutions.
- Created a reproducible pipeline for data science, from data transformation to hyper-parameter model tuning.
- Placed a special emphasis on model interpretability.
Technologies: Scikit-learn, Jupyter, Python
Data Scientist | Machine Learning Engineer
2016 - 2019
Nordeus
- Created a neural network model to generate football player faces in a scalable way. The outputs from this model are used in one of the company's games.
- Designed matchmaking algorithms in the Top Eleven game, a soccer manager simulation with over 200 million users worldwide using game theory and Monte Carlo techniques.
- Worked with the internal customer support team to automate the process of tagging player complaints using NLP techniques.
- Developed a predictive model to estimate the marketing campaigns' ROAS (return on ad spend).
- Managed two junior data scientists responsible for business intelligence and game system design.
Technologies: Scikit-learn, Tableau, Impala, Hadoop, Python, Data Analysis
Quantitative Risk Analyst
2012 - 2016
Erste Group Bank
- Implemented and validated in MATLAB and Python all models used by Erste Group Bank to price and hedge interest rate derivatives.
- Wrote exhaustive documentation for each validated model to present to the European Central Bank.
- Proposed and implemented improvements to the methodology used to estimate the credit market risk of the banking and trading books.
- Backtested the performance of different value at risk models to propose improvements to the methodology used by the bank.
- Mentored junior quantitative risk analysts that joined the team.
Technologies: MATLAB, Python
Experience
Realistic Human Face Generator for Mobile App Golden Boot 2019
The problem I wanted to solve was to fully automate the process of generating profile images of football players for several of the company's games. The system is used in the mobile game Golden Boot 2019, available in both iOS and Android, and with over one million installs since its release.
I built a pipeline of three models, each applied sequentially. First, a cutting-edge GAN network retrained to my own dataset that generates realistically looking football player faces. Second, a logistic classifier built from the last layer of a VGG network, to classify the output of the GAN into "good" faces and "bad" faces, ensuring that only images of sufficient quality are displayed to the user. Third, another logistic regression on top of the last layer of a VGG net to classify the face according to its ethnicity. This last step was necessary in order to have control over the correlation between football player nationality and his physical appearance.
I built a pipeline of three models, each applied sequentially. First, a cutting-edge GAN network retrained to my own dataset that generates realistically looking football player faces. Second, a logistic classifier built from the last layer of a VGG network, to classify the output of the GAN into "good" faces and "bad" faces, ensuring that only images of sufficient quality are displayed to the user. Third, another logistic regression on top of the last layer of a VGG net to classify the face according to its ethnicity. This last step was necessary in order to have control over the correlation between football player nationality and his physical appearance.
Customer Support Automatization with Natural Language Processing
A natural language processing system to automatically classify customer issues. The tool was developed during my time at Nordeus, a mobile gaming company with over two million daily active users. The end-goal was to reduce the number of tickets that human agents had to process, and increase customer satisfaction overall.
Financial Markets Valuation and Explanation Using Machine Learning
A suite of machine learning models to quantify and explain the valuations of banking institutions in the Gulf region. I was the machine learning engineer in charge of designing and implementing such a system according to the data availability and end goals of the client.
Fleet Optimization and Demand Forecasting with Gradient Boosting
Lola Market is a Spanish startup that lets you order groceries online at your favorite shop and get them delivered to your home. The company has a fleet of shoppers that go to the stores, do the shopping, and take it to the user's house.
A big question for Lola's operations team was: "how many shoppers should be available at each location and hour of the day to guarantee 100% availability to our users and to minimize shopper idle hours?". The goal of the project was to automate and improve the allocation of shoppers in geographies and timeslot.
The solution I developed is a machine learning (ML) model that predicts future user demand at each geography (city, district) and hour of the day for the following two weeks. I also developed a suite of Tableau dashboards to make the system transparent to Lola's Operations team.
A big question for Lola's operations team was: "how many shoppers should be available at each location and hour of the day to guarantee 100% availability to our users and to minimize shopper idle hours?". The goal of the project was to automate and improve the allocation of shoppers in geographies and timeslot.
The solution I developed is a machine learning (ML) model that predicts future user demand at each geography (city, district) and hour of the day for the following two weeks. I also developed a suite of Tableau dashboards to make the system transparent to Lola's Operations team.
The Hands-on Reinforcement Learning Course
https://datamachines.xyz/the-hands-on-reinforcement-learning-course-page/Created and published an online course on reinforcement learning (RL), from fundamentals to cutting-edge deep RL.
The course is available online for free.
The goal is to teach my students, with a hands-on approach, how to implement the key RL algorithms from scratch using Python and PyTorch.
The course is available online for free.
The goal is to teach my students, with a hands-on approach, how to implement the key RL algorithms from scratch using Python and PyTorch.
Hands-on LLM Course
https://www.realworldml.net/the-hands-on-llm-courseA hands-on course to learn how to build and deploy large language model (LLM)-powered applications.
In this tutorial, you will design, build, and deploy a financial advisor using LLMs and MLOps best practices. This hands-on tutorial will help you go beyond LangChain demos in Jupyter notebooks and build real-world machine learning (ML) products using LLMs.
In this tutorial, you will design, build, and deploy a financial advisor using LLMs and MLOps best practices. This hands-on tutorial will help you go beyond LangChain demos in Jupyter notebooks and build real-world machine learning (ML) products using LLMs.
Real-time Machine Learning System
A live, interactive course in which we (you and me) will build a modular real-time machine learning (ML) system to predict short-term crypto prices.
This course is ideal for those who are familiar with Python and want to learn how to build modular ML systems that transform raw data into predictions in real time.
This course is ideal for those who are familiar with Python and want to learn how to build modular ML systems that transform raw data into predictions in real time.
Education
2011 - 2012
Master's Degree in Quantitative Economics
Ca'Foscari University Venice - Venice, Italy
2010 - 2011
Master's Degree in Quantitative Economics
University of Bielefeld - Bielefeld, Germany
2005 - 2010
Master's Degree in Mathematics
Polytechnic University of Catalonia - Barcelona, Spain
Certifications
JULY 2005 - PRESENT
Participant in the 46th International Mathematical Olympiad
International Mathematical Olympiad
Skills
Libraries/APIs
Scikit-learn, Keras, TensorFlow, OpenCV, PySpark, REST APIs, Shapely, XGBoost, PyTorch
Tools
Tableau, PyCharm, Impala, MATLAB, Jupyter
Languages
Python, SQL, C++, Python 3
Frameworks
Flask, Hadoop
Platforms
Google Cloud Platform (GCP), Amazon Web Services (AWS), MacOS
Storage
Redshift
Other
Machine Learning, Natural Language Processing (NLP), Statistics, Statistical Modeling, Computer Vision, Data Science, Quantitative Finance, Mathematical Modeling, Data Visualization, Generative Pre-trained Transformers (GPT), Large Language Models (LLMs), Time Series Analysis, Deep Learning, AWS Database Migration Service (DMS), Explainable Artificial Intelligence (XAI), Data Engineering, Segment, Random Forests, Mathematics, Optimization, Genomics, Custom BERT, DeepAR, Deep Reinforcement Learning, Reinforcement Learning, Data Analysis, Real-time Data, Quantitative Modeling
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring