Skip to content
View marianotorres-code's full-sized avatar

Block or report marianotorres-code

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
marianotorres-code/README.md
Typing SVG

🚀 About me

  • 🔭 Currently building Vatia — a simulation and optimization tool for solar + battery storage sizing at industrial plants in northern Mexico, tackling grid instability from the local utility.
  • 🧠 Into startups and building — I'd rather ship something small and real than plan something big and theoretical.
  • 🎓 Heading toward a degree in AI Engineering.
  • 📊 I write up my ML projects with an emphasis on honest validation — every reported improvement gets checked against noise before I trust it, and every leaderboard gap gets diagnosed with numbers, not guesses.
  • 💬 Ask me about feature engineering gone wrong, leakage-safe validation, or why your CV score doesn't match your leaderboard score.

🛠️ Stack

Python FastAPI Next.js Supabase scikit-learn Pandas


📌 Featured projects

titanic-ml — 0.79665 on the public leaderboard. Five improvement attempts tried, one confirmed real (leave-one-out group survival rate) after ruling out feature engineering, hyperparameter tuning, and ensembling as noise.

houseprices-ml — Top 8.5% (310/3655) without external data leakage. Diagnosed a CV-to-leaderboard gap down to a single mispriced outlier — then found the diagnostic metric itself had a blind spot and corrected it on the record.


📊 GitHub stats


Building things, shipping them, and writing down when I was wrong.

Popular repositories Loading

  1. titanic-ml titanic-ml Public

    Titanic survival prediction — feature engineering, leakage-safe validation, and honest model comparison

    1

  2. Throw Throw Public

  3. houseprices-ml houseprices-ml Public

    Kaggle House Prices — top 8.5% (310/3655) without external data leakage. LOO-safe target encoding, honest CV, and a leaderboard-gap diagnosis that found a single mispriced outlier.

    Python

  4. marianotorres-code marianotorres-code Public

  5. digit-recognizer-ml digit-recognizer-ml Public

    CNN pipeline for the Kaggle "Digit Recognizer" competition (MNIST)

    Python

  6. autograd-from-scratch autograd-from-scratch Public

    A reverse-mode automatic differentiation engine built from zero dependencies (no PyTorch, no TensorFlow — just Python and the chain rule)

    Python