MKDownload CV
06 /

ML from scratch

Builder — university project · Sep 2025 — Jan 2026

Machine learning built from raw mathematics in Java — no AI libraries allowed — predicting student grades from 16,994 real records.

AI engineeringData analysis

Anyone can import a machine-learning library. Writing one is different: every tree of a 50-tree Random Forest, the feature engineering, the statistics of the evaluation — all implemented by hand in Java. On 16,994 real student grades it beat the baseline by 61%, with 95% of predictions landing within ±1.16 grade points. Plus an interactive dashboard with 16+ visualisations to explore the results.

  • Zero machine-learning libraries — the entire algorithm written from mathematical first principles
  • 61% more accurate than baseline on 16,994 real student grades
  • 95% of predictions within ±1.16 grade points

16,994

real records

+61%

over baseline

0

ML libraries used

Java · JavaFX · Mathematics · Statistics