DATA SCIENTIST · ML RESEARCH · AI ENGINEER

ML Research
& AI Engineering

Data Science student at the Polytechnic University of Valencia, with multiple honors for academic excellence. I’m passionate about mathematics and artificial intelligence, with special interest in their application in consulting and finance.

Experience in AI applications for wealth management at a leading Spanish bank and in machine learning research at the PRHLT Research Center.

  • Data science
  • ML research
  • AI engineering
Portrait of Pablo Gandía
BSc Data Science, Engineering Universitat Politècnica de València (UPV)
  • Many honor distinctions for academic excellence.
  • Top academic percentile in the cohort.
Sigma Data Club UPV logo
Sigma Data Club UPV

Former Board Member

UPV Investment Club logo
UPV Investment Club

Software Member & Analyst

Computer Science & Technology Exchange Beihang University · Beijing
  • Top 10 engineering university in China.

Focus

  • RLReinforcement Learning
  • AIAdvanced Artificial Intelligence

SELECTED WORK

Projects

AI engineering and data systems projects across model evaluation, multimodal reasoning, APIs, data pipelines and applied machine learning.

Error estimation PRHLT logo
confidence bins error
LM Nelder-Mead Temperature scaling
Scores Calibrate Metrics
ML research AERFAI 2026

Unsupervised Error Prediction

A command-line toolkit that estimates classifier error from confidence scores, calibrates predictions with power-law fitting or temperature scaling, and exports reproducible metrics, plots and experiment metadata.

  • Levenberg-Marquardt, Nelder-Mead and temperature scaling strategies.
  • Grouped evaluation, optional outlier trimming and seven calibration metrics.
  • Presented as a practical toolkit at the AERFAI 2026 Summer School.
Visual mathematics Qwen2.5-VL
Advanced geometry and algebra problem from the MathVision project
  1. OCR + classifyProblem image
  2. Crop evidenceStatement and options
  3. Qwen reasoningA-E analysis
  4. Answer + chatTraceable solution
Multimodal research Visual reasoning

MathVision AI

An auditable research and educational prototype for visual multiple-choice mathematics. It classifies the problem, extracts text and visual regions, analyses options A-E separately, and exposes the evidence behind the final answer.

  • Local multimodal pipeline with OCR, visual classification and crop inspection.
  • Option-by-option reasoning, visible traces and follow-up chat in four languages.
  • FastAPI application with local Ollama inference and Qwen-based reasoning.
NewsReader AI political news analysis interface
NLPLLMs

Political News Bias Analysis

A Streamlit application that turns an article URL into political-bias probabilities, real/fake scores and a Mistral-generated explanation using fine-tuned BERT and RoBERTa models.

  • Fallback scraping with requests, Trafilatura and Selenium.
  • Fine-tuned on 63k+ credibility and 10k+ bias articles.
Traffic data analysis in Valencia
Data analyticsMobility

Valencia Traffic Analysis

An urban-data study of 1,129 Valencia traffic sensors, transforming five-minute measurements into hourly patterns and relating mobility to air quality and wind conditions.

  • Temporal analysis, PCA and k-means road segmentation.
  • Highest-scoring project in the course: 9.9/10.
Tastytrade API integration project
Data engineeringFintech

Tastytrade API

A lightweight Python wrapper that unifies Tastytrade REST workflows and DXFeed WebSocket streaming for market data, account analytics and order execution.

  • Authentication, positions, transactions and orders.
  • Real-time and historical market data in Python-ready workflows.
Real-time facial emotion detection demonstration
Computer visionDeep learning

Facial Emotion Detection

A real-time facial-expression classifier covering seven emotions, trained on FER2013 with Keras and deployed through an OpenCV and Tkinter desktop application.

  • 35,887 images with augmentation and class weighting.
  • CNN with dropout, early stopping and live webcam inference.

STACK

Tools I use to build

Focused on practical data science workflows: experimentation, model evaluation, APIs, data processing and reproducible engineering.

Python PyTorch TensorFlow scikit-learn FastAPI Flask SQL Databricks AWS Docker Git Cloud ML APIs R pablo.gandia.minana@gmail.com