Binary sentiment classifier for IMDB movie reviews, built with scikit-learn (TF-IDF + Logistic Regression, 91.4% test accuracy).

🇧🇷 marolaGPT + the fine-tune, the DSPy prompt compile, the local model image and the benchmark gate

Hybrid Network Intrusion Detection and Static Malware Analysis Platform built with Python, Streamlit, Scapy, and Machine Learning.

An AI-powered automotive mobile advisory app featuring ML-driven car resale value prediction, intelligent recommendation engine, and interactive marketplace built with React Native, Expo, Flask, and Firebase.

Two labels. 577 trainable parameters. Wildly unreasonable startup ambition. A Silicon Valley-inspired computer vision training lab in Mojo.

Short, simple controller games designed for seniors that use an ML model to flag early signs of movement disorders.

LLM-agnostic agent harness for trading futures on Claude Code, Codex, or Qwen Code, with a risk-gated order path and pluggable broker MCP servers (TopstepX default).

Which MLB players are underpaid or overpaid? Random-forest models predict each player's salary from his season stats, 2016–2026, and compare it to what he's actually paid.

A deep learning project that trains neural networks to classify images into categories using computer vision techniques and image datasets.

My portfolio

Cost-sensitive IEEE-CIS fraud detection with a LightGBM pipeline and an interactive Streamlit policy simulator.

Controlled comparison of Random Forest vs SVM for phishing URL detection (PhiUSIIL)

AI-Driven Cyber Threat Prediction and Real-Time SOC Intelligence System using Ensemble ML and MITRE ATT&CK

FastAPI + React/TypeScript app serving four ML models: pneumonia X-ray, brain-tumor MRI, Parkinson's voice and stroke risk

Kinship reasoning without a neural network: a sound, set-valued model that derives its rules from opaque symbols. CLUTRR, all six datasets.

Sensor-missingness forecasting notebook with 16 availability masks, core/specialist routing and data-free smoke checks.

DeepKit-ml A lightweight machine learning library built from scratch in Python, implementing core algorithms and optimization techniques.

《机器学习中的数学》(Mathematics for Machine Learning) 中文版 LaTeX 工程与 408 页编译成品 PDF | 基于 dxxzst/mml-book-chinese 译本整合构建

⚡ AI & Tech Explorer | Python Developer 🤖 Building Telegram Bots & Smart Automations 💻 Exploring Python, APIs & Open Source 🚀 Learn • Build • Innovate • Grow 🌌 Turning Ideas Into Digital Reality 🔮 Dream Big. Code Smart. Stay Limitless.

Watch a language model think: next-word predictions and live attention, visualized as a galaxy — runs 100% in your browser.

Phishing email analyzer: SPF/DKIM/DMARC, spoofing, deceptive links, risky attachments + NLP classifier fused into an explained 0-100 risk score. Recall 100%, ROC-AUC 0.998.

Kaggle: verify article contributors with TF-IDF/SVD embeddings, author profiles and leakage-free XGBoost (score 0.88)

Budgeted counterfactual debugging for ML pipeline regressions, including interacting changes across models, data, preprocessing, configs, and environments.

Production-style customer churn prediction system: reproducible DVC pipeline, MLflow-tracked comparison of Logistic Regression, Random Forest and XGBoost (best ROC-AUC 0.842), Airflow orchestration, Dockerized FastAPI inference, and Gemini-generated retention offers for at-risk customers.

Telecom enterprise churn prediction and customer lifetime value (LTV) forecasting pipeline on 10M+ call records (0.88 AUC).

nvidia-smi for multiple servers: every NVIDIA GPU and who is using it, in one terminal over SSH. Nothing to install on the servers.

Practice Projects of Machine Learning.....

Credit Card Fraud Detection using Machine Learning, Random Forest, SMOTE, and an interactive Streamlit dashboard to identify potentially fraudulent transactions.

GSoC organization analytics tool that scrapes historical participation data, exports structured Excel datasets, and predicts next-year organization presence using machine learning.

Reference-assisted evaluation of a granular-computing matting pipeline for mammogram masses (CBIS-DDSM, MIAS), with seed-only controls and paired bootstrap statistics.