Python framework for LLM-driven weight-space programming. Construct, evaluate, and export neural networks with PyTorch, Ollama, or OpenRouter.
-
Updated
Nov 4, 2026 - Python
Python framework for LLM-driven weight-space programming. Construct, evaluate, and export neural networks with PyTorch, Ollama, or OpenRouter.
1st place in sch2-reg-2026-d4-1 (ROC AUC 1.0) via recovered test labels, plus an honest modelling study on playground-series-s5e3. Documents the structural probes proving the public LB grades only 146 of 730 test rows.
k-Nearest Neighbours (scikit-learn), for PhiUSIIL phishing URL classification. 500,002 stored values, served over HTTP.
The IF3070 Foundations of Artificial Intelligence (STEI ITB, 2024/2025-1) coursework split of the PhiUSIIL Phishing URL Dataset — 140,404 labelled rows and a 10,000-row unlabelled holdout, CC BY 4.0. Mirrored on Hugging Face.
k-Nearest Neighbours implemented from scratch, for PhiUSIIL phishing URL classification. 500,001 stored values in a NumPy .npz, no pickle, served over HTTP.
Weather Station Anomaly Detection
PyTorch Random Fourier Features classifier for white wine quality prediction with 3-class and binary experiments.
An experimental supervised ML loop with saved-model evaluation, optional expert review, and a technical report on its results and limits.
Auditable clinical tabular benchmark for hospital and measurement-policy shift, with reconstructable evidence and explicit claim boundaries.
🤖 Lead.AI Labs — Hugging Face AI Portfolio | Model Cards, Datasets, Demo Spaces for Fraud Detection, Customer Intelligence & NLP
Supervised binary classification, random forest classifier, feature engineering, feature scaling, probability calibration, tabular data inference
End-to-end AutoML platform for tabular classification with model versioning and live FastAPI deployment.
Gaussian Naive Bayes implemented from scratch, for PhiUSIIL phishing URL classification. 198 parameters in plain JSON, no pickle, served over HTTP.
Gaussian Naive Bayes (scikit-learn), for PhiUSIIL phishing URL classification. 202 parameters, served over HTTP.
To associate your repository with the tabular-classification topic, visit your repo's landing page and select "manage topics."