Artificial Intelligence Engineer with experience developing end-to-end solutions following the CRISP-DM methodology, working from requirements gathering and communication with multidisciplinary teams to data collection, processing, and modeling using SQL and Python.

Experienced in Data Science, performing exploratory data analysis, data visualization, dashboard creation, predictive analytics, and the development of customized Machine Learning solutions focused on cost reduction, churn prevention, fraud detection, and data-driven decision-making.

My academic background, combined with experience in Research and Development (R&D) projects, provided me with strong analytical skills, structured logical thinking, and the ability to deeply investigate complex problems involving Artificial Intelligence, GenAI/LLM, NLP, Recommendation Systems, and Computer Vision.

I am communicative and enthusiastic about knowledge sharing. In addition to published scientific papers, I have participated in podcasts, interviews, and produce AI-related technical content on Medium.

• Email: italodepontesoliveira93@gmail.com
• LinkedIn: linkedin.com/in/italo-de-pontes
• GitHub: https://github.com/italoPontes/italoPontes
• Substack: https://substack.com/@italodepontes
• Google Scholar: https://scholar.google.com/citations?user=3R5wdZIAAAAJ

• Performed large-scale exploratory data analysis and statistical correlation studies on transactional datasets to design The Smart-Monitor, an intelligent fraud prevention platform focused on device fingerprinting, behavioral analytics, and device-level risk signals.
• Reduced manual review SLA through more accurate prioritization and automated risk enrichment, improving operational efficiency and strengthening product trust, safety, and market sophistication.
• Developed behavioral and recurrence-detection strategies capable of identifying systematic attackers across multiple transactions and devices, increasing recall by 13% while reducing false positive rates by %9.

• Led end-to-end AI and Data Science projects following the CRISP-DM methodology, conducting pre-sales technical discussions, stakeholder alignment, solution architecture, deployment, and technical reporting for enterprise clients.
• Delivered Machine Learning and Analytics solutions for companies including Heineken, Gerdau, Leroy Merlin, B3, and the Brazilian Ministry of Education.
• Developed regression models for Heineken to predict the optimal beer fermentation point in industrial tanks, reducing unnecessary laboratory quality tests and optimizing tank retention time for more efficient production reuse.
• Developed predictive models for Gerdau to estimate the calorific efficiency of coke generated from different charcoal blend compositions, supporting energy optimization and reducing operational composition costs in industrial processes.
• Built anomaly detection and recommendation systems for Leroy Merlin to identify inconsistent SAP enterprise records and suggest corrected values for affected fields, reducing operational inconsistencies and customer-related issues.

• Developed churn prediction models for B2B medical equipment maintenance contracts, performing feature engineering, exploratory analysis, and correlation studies using operational, financial, and contractual variables, while building daily predictive pipelines for customer retention monitoring.
• Enabled early identification of contracts with high churn propensity up to three months in advance, achieving approximately 35% predictive precision and generating a return on investment that paid back the project in less than one year.

• Developed NLP solutions for automated customer support email responses in a large-scale enterprise environment handling more than 1 million emails per month. • Built hierarchical classification and recommendation systems for automatic support request categorization and response template ranking. • Improved operator productivity and reduced SLA response time through intelligent automation, without negatively impacting customer satisfaction metrics. • Implemented sensitive data anonymization pipelines for privacy protection, increasing security against internal data leakage by customer support operators.

• Developed Computer Vision and Deep Learning solutions for 3D image as part of the HP Sprout ecosystem, aiming to enhance brand exposure and improve user experience through intelligent object recognition services.
• Developed age and gender estimation models using convolutional neural networks and a proprietary Data Augmentation methodology proposed during my master’s research, achieving superior performance in 7% compared to Google Inception v3, considered state-of-the-art at the time.
• Improved facial detection accuracy using skin detection post-processing techniques.

• Contributed to software engineering research focused on UML and model-driven development by expanding ATL Analyzer functionalities and developing Java-based structures for automated Ecore metamodel transformations.
• Improved the scalability and operational efficiency of model transformation workflows, reducing processing complexity from exponential-like growth behavior to a linear.

Recommender Systems (2017-2020)

Federal University of Campina Grande (UFCG)

Computer Vision & Deep Learning (2014-2016)

Federal University of Campina Grande (UFCG)

Digital Video Processing (2010-2013)

Federal Institute of Education, Science and Technology of Paraíba (IFPB)

• Languages: Python, SQL, Shell Script, C/C++.
• GenAI: LLMs, Embeddings, Vector Search, Prompt Engineering, RAG.
• Machine Learning: Scikit-Learn, CatBoost, TensorFlow/Keras, PyOD.
• Data Engineering: PySpark, Pandas, Numpy, SciPy, Imbalance-learn.
• Computer Vision: OpenCV, YOLO, Caffe, Scikit-Image.
• Visualization & Analytics: Seaborn, Streamlit, Plotly, Matplotlib, Ydata-profiling, Missingno.
• NLP: HuggingFaces, NLTK, Gensim, Microsoft Presidio, Regex.
• Recommendation Systems: Matrix Factorization, LightFM (Factorization Machines), Transformers4Rec.
• Interpretability: SHAP, LIME, What-if.

• Developed a web-based steganography platform for hiding and extracting secret messages inside digital images.
• steganography.streamlit.app

Python, Streamlit, OpenCV, NumPy

• Developed a Machine Learning pipeline for fraudulent financial transaction detection, achieving a Top 70 (<3.5%) ranking among more than 2,000 submissions in a Zindi Challenge competition.
• github.com/fernandojunior/financial-fraud-detection

Python, Scikit-Learn, PySpark, CatBoost, Pandas, PyOD

• Video Audience Analysis using Bayesian Networks and Face Demographics Conference on Graphics, Patterns and Images (SIBGRAPI - 2019)

• A data augmentation methodology to improve age estimation using convolutional neural networks Conference on Graphics, Patterns and Images (SIBGRAPI - 2016)

📎 Full publication list available on https://scholar.google.com/citations?user=3R5wdZIAAAAJ.

• Software Patent – SQUALES Registration Number: BR 51 2014 000292-6.

Available at: drive.google.com/file/d/1hgIszxpRKthN99LEtSK9V5koOICh7h6s

• Substack: https://substack.com/@italodepontes

• IEEE Young Professional
• Former Chair of the IEEE Student Branch at IFPB-CG
• Student Member of the Brazilian Telecommunications Society (SBrT)
• Podcast participant discussing AI, Data Science, and innovation
• Technical content creator focused on simplifying complex AI topics