Iker Pacheco

Watch video

DemoVideo

EndoTector

Polyp segmentation in endoscopy with YOLO-seg

Median Dice

0.96

mAP@50

0.95

Real-time inference.

Problem

Outline polyps in endoscopy images with a model light enough to run in real time.

Approach

Fine-tuned a 6 MB YOLO-seg model on the Kvasir-SEG dataset, with an interactive Gradio web demo.

Stack

• Python
• YOLO-seg
• OpenCV
• Gradio
• Docker
• GitHub

CAI-listhenics

Pull-up technique analysis with computer vision

End-to-end system deployed: Python backend and Gradio client on Hugging Face Spaces.

Problem

One day, training back, I realised how hard it is to count reps and judge the technique of every pull-up by eye. An AI support system that measures each rep and returns objective exercise metrics adds exactly that value.

Approach

Pose estimation with MediaPipe and per-rep kinematics, such as the eccentric/concentric ratio, computed with OpenCV and SciPy.

Stack

• Python
• MediaPipe
• OpenCV
• SciPy
• Gradio
• Hugging Face Spaces

TalkPoster

Early ALS detection from voice

Explainable ML on acoustic biomarkers from smartphone recordings

Bachelor's thesis, graded 10/10. Nominated for CASEIB 2026.

AUC

0.831

Sensitivity

80%

F1

81.2%

LASSO + XGBoost validated with 10-fold CV and 95% CIs.

Problem

Detect ALS early from smartphone voice recordings, with an interpretable model.

Approach

Explainable ML pipeline on acoustic biomarkers: LASSO feature selection, XGBoost classification and SHAP explanations.

Stack

• Python
• scikit-learn
• Praat-Parselmouth
• Explainable AI (XAI)
• Statistics
• SciPy
• Signal Processing
• Data Science
• Data Engineering