Machine Learning
scikit-learnXGBoostModel evaluationFeature engineeringSupervised learningUnsupervised learningAutosupervised learning
AI Engineer Portfolio
Buenos Aires, Argentina
Santiago Tomas Torres
AI & ML Engineer | Data Scientist
Final-year Artificial Intelligence Engineering student building machine learning, computer vision, and LLM systems across research collaborations and applied product work.
Machine LearningComputer VisionNLP / LLM Systems
Thesis
Ongoing research
Efficient Continual Learning on Medical Foundational Models: Analysis of Adaptation Strategies for Medical Image Segmentation
Current undergraduate thesis work within the B.Eng. in Artificial Intelligence Engineering at University of San Andres.
Research status
In ProgressExploring efficiency trade-offs of continual learning strategies for adapting foundational models like SAM to medical image segmentation tasks, with a focus on practical implications for real-world deployment.
Skills
Core technical stack
Deep Learning
PyTorchTensorFlowPEFT / LoRATime-series modeling
NLP / LLMs
OpenAI APIAzure OpenAIOllamaspaCyNLTKRAGPrompt engineeringLangGraph
Computer Vision
OpenCVYOLOCNNsDiffusion modelsFlow-based models
Reinforcement Learning
Policy GradientActor-criticSoft Actor CriticDQN/DDQNDDPG
MLOps / Backend
MLFlowFastAPIDockerDocker ComposeStreamlitKubernetesCI/CD pipelines
Cloud / DevOps
LinuxAWSAzureGitGitHub Actions
Data / Databases
PostgreSQLMongoDBApache SparkNumPyPandasMatplotlib
Projects
Selected AI projects
A mix of local LLM systems, NLP workflows, computer vision experiments, and supervised learning projects drawn directly from the CVs.
Personal projectJan 2026
Text-to-SQL Assistant with Local LLMs
Designed and implemented an end-to-end Text-to-SQL pipeline that converts natural language queries into executable SQL and narrative answers over PostgreSQL sales data.
Added schema-aware prompting, a two-stage inference flow, self-correction, and read-only guardrails to improve reliability in smaller local models.
View GitHub repoPythonFastAPISQLAlchemyPostgreSQLOllamaDocker Compose
University of San AndresJun 2025
Soccer Commentator
Built a low-latency workflow for real-time soccer commentary generation with NLP, text-to-speech, and distributed microservices.
Integrated LLM and speech components with gRPC and FastAPI, including LoRA-based voice customization for the TTS model.
View GitHub repoOpenAI APIAzure OpenAIOllamaFastAPIWhispergRPCDocker
Reinforcement learning projectDec 2025
Soft Actor-Critic for Donkey Car Gym
Implemented Soft Actor-Critic from scratch and evaluated its behavior in the Donkey Car Gym environment.
Focused on policy-learning stability, continuous-control experimentation, and practical evaluation in a simulated autonomous driving task.
View GitHub repoReinforcement LearningSoft Actor-CriticDonkey Car GymContinuous control
University of San AndresDec 2024
Wildlife Monitoring with Thermal Images
Automated wild-animal detection from RGB and thermal drone imagery using multi-modal object detection models.
Explored computer vision approaches for aerial monitoring using YOLOv5 and YOLOv11 baselines.
Computer VisionYOLOv5YOLOv11Deep Learning
University of San AndresJun 2024
Prediction of SUV Prices on Mercado Libre
Trained and compared supervised regression models to estimate SUV prices from Mercado Libre listings.
Combined preprocessing, feature engineering, and regression metrics to establish a practical predictive baseline.
Pythonscikit-learnXGBoostMachine Learning
Experience
Research collaborations and applied experimentation
Nov 2024 - Jan 2026
Research Team Member
University of San Andres × Northwestern University
San Fernando, Argentina
PythonLLMsOpenAI fine-tuningWeb scrapingData labeling
- Co-authored the resulting publication, Linnaeus: A Hierarchical, Multi-Label Framework for Autonomous System Classification.
- Designed and executed a data pipeline covering web scraping, cleaning, and labeling for Internet Autonomous System classification.
- Fine-tuned LLMs for multiclass classification and evaluated performance to iterate on data and modeling strategy.
Mar 2025 - Dec 2025
Research Team Member
University of San Andres
San Fernando, Argentina
Nvidia Isaac SimROS2Synthetic dataYOLO
- Generated a photorealistic synthetic dataset with automated capture and labeling for future YOLO training.
- Combined Nvidia Isaac Sim, ROS2, and simulated drone navigation to support computer vision experimentation.
Feb 2024 - Dec 2024
Research Team Member
Interdisciplinary Laboratory of Time and Experience (LITERA) - UdeSA
San Fernando, Argentina
PythonMachine LearningDeep LearningData collection
- Built models to predict sleep-event onset and offset at both individual and population levels.
- Worked with Android activity time-series data and validated results against wearable actigraphy ground truth.
Research
Publications
arXiv preprint [cs.NI] • Mar 2026
Linnaeus: A Hierarchical, Multi-Label Framework for Autonomous System Classification
Piotto, M.; Schuemer, I.; Torres, S. T.; Beiro, M. G.; Carisimo, E.; Bustamante, F. E.
Achievements
Awards and recognitions
Oct 2025
1st Place Hackathon
Siemens, Louis Dreyfus Company, and Argentine Industrial Union
Optimization proposal for the soybean grain sampling process with a computer vision and machine learning focus.
Mar 2025
1st Place HackITBA - Sustainability Category
HackITBA
Built an NLP-driven platform to analyze and compare unstructured ESG reports across companies.
Mar 2024
2nd Place - AI E-Commerce Challenge
AWS, Trafilea, and Mutt Data
Contributed to a generative AI concept for personalized ads based on user data.
Education
Academic foundation
Mar 2022 - Present
B.Eng. in Artificial Intelligence Engineering
University of San Andres
San Fernando, Buenos Aires, Argentina
- Student in the first cohort of the country's first Artificial Intelligence Engineering program.
- Currently in the final year of the degree.
- Active thesis work in progress.
Contact Me
Feel free to reach out through any of these channels.
For AI engineering, applied ML, computer vision, or research-oriented opportunities, email is the fastest way to reach me.
