Browse categories
Explore
Fiverr Pro
English
$
USD
I will audit and debug your machine learning pipeline for hidden errors
Fernando Cajar
About this gig
Is your ML model underperforming, overfitting, or giving results that don't quite add up?
I audit machine learning pipelines to find hidden issues that quietly ruin model performance: data leakage, incorrect train/validation splits, overfitting, poor evaluation metric choices, and biased sampling.
I recently identified and fixed a critical data leakage issue in a medical imaging classification project (image-level vs. patient-level splitting), which significantly changed the validated results. This is the kind of error that's easy to miss and hard to detect without a careful review.
What I check:
- Data splitting methodology (leakage risks)
- Class balance and sampling strategy
- Evaluation metrics fit for your problem
- Overfitting and underfitting signals
- Model architecture fit for your data type
You'll receive a clear written report explaining what I found, why it matters, and how to fix it.
Let's make sure your model results are actually trustworthy.
Get to know Fernando Cajar
Fernando Cajar
ML and Data Analyst Medical Imaging Specialist
- FromPanama
- Member sinceJul 2026
Languages
Spanish, English
Biomedical Engineer with hands-on experience in machine learning applied to medical imaging. I led a government-funded research project (SENACYT, Panama), building a hierarchical deep learning system for breast cancer detection using CNN architectures, achieving strong diagnostic accuracy across multiple datasets.
I specialize in: image classification models, ML pipeline auditing (data leakage, overfitting, validation errors), statistical analysis, and technical documentation (LaTeX/IEEE).
I also work as a data analyst, producing survey analysis and reports.
FAQ
What do you need from me for the Basic package?
Your data splitting script, plus the file with your dataset metadata (IDs, groups, labels). This lets me check if related samples, like multiple images from the same patient, subject, or entity, are properly kept together between train and validation sets.
What if you don't find any issues?
You'll still get a written report confirming what was reviewed and why it looks solid. That confirmation has value too, it means you can trust your results.
Do you work with medical imaging data specifically?
Yes, that's one of my areas of focus, but I also work with general computer vision and tabular ML pipelines across other domains.
Can you review pipelines built in TensorFlow, not just PyTorch?
Yes, my core experience is in PyTorch, but I'm comfortable reviewing TensorFlow/Keras pipelines as well.
