Selectividad Mates Tierlist

Comparing the difficulty of 297 real Spanish university entrance exams (EvAU) across 16 years and 17 regions.

  1. Built the entire data pipeline with zero lines of code, using markdown files instead of scripts (AI-native software engineering).
  2. Ran parallel AI agents to parse PDF exams into structured JSON and rate 2,288 questions against a unified rubric.

Results show that exam difficulty varies significantly depending on the region, but has remained stable over time.

Selectividad Mates Tierlist Dashboard

GPT Primavera

Benchmarking math olympiad problems using local and cloud LLMs.

  1. Reverse engineered concursoprimavera.es API to download 160+ exams programmatically.
  2. Converted PDF exams to JSON questions using Gemini API (including image descriptions).
  3. Benchmarked accuracy, token usage and time to solve each question.

Results show cloud models are still superior in intelligence (96% vs 45% accuracy), speed (10x faster), and token-efficiency than small (<9b) open source models.

GPT Primavera Project Screenshot

Getcanvas

Python Terminal user interface (TUI) to download course materials from Canvas LMS. Published package to PyPI.

Getcanvas Python Package

Brainport Jobs

Static frontend aggrupating job opportunities in the Brainport region (Eindhoven, Netherlands). Uses GitHub Actions to scrape jobs daily from brainporteindhoven.com.

Brainport Jobs Dashboard

UFC Predictions

Predicting UFC fight outcomes using historical fighter data.

  1. Daily scraping of ufcstats.com using GitHub Actions.
  2. Trained ensemble of models (XGBoost, RandomForest, etc.) and pushed them to Huggingface.
  3. Daily inference using GitHub Actions + Huggingface Inference API to predict upcoming fights.
UFC Predictions Dashboard