about_me:
role: "Machine Learning & AI Enthusiast"
focus: "Real-world problem solving over toy demos"
strengths:
- Data analysis, feature engineering, model optimization
- Deep Learning (CNNs), NLP, Computer Vision
- Explainable AI (LIME, SHAP)
language: Python
mindset: "Learn by building. Ship clean code. Iterate."🧠 FER_XAI — Explainable Facial Expression Recognition
7-class facial expression recognition with post-hoc explainability via LIME and SHAP. Two CNN backbones — ResNet-50 and EfficientNet-B0 — trained and benchmarked on FER2013 and RAF-DB, with pixel- and region-level explanations generated for every single prediction.
- 🏆 Best result: 84.36% accuracy (ResNet-50 on RAF-DB)
- 🔍 Every prediction comes with a visual "why" — not just a label
- 🛠️ Stack:
PyTorch·LIME·SHAP·AMP training
🤖 CarterX — Full-Stack Intelligent Application
A full-stack application pairing a Python backend with a JavaScript frontend, backed by its own local database layer for structured data handling — built with a dedicated test suite for reliability.
- ⚙️ Split
app/+frontend/architecture - 🧪 Includes its own test suite
- 🛠️ Stack:
Python·JavaScript·HTML/CSS·SQLite
🎣 Phishing-Detection — Real-Time URL Threat Classifier
An ML-powered system that extracts 30+ security-relevant features from any URL — structural, content-based, and domain-based — to classify it as legitimate or malicious in real time. Fully containerized and deployed on AWS via an automated CI/CD pipeline.
- ⚡ ~95%+ accuracy on held-out test data
- ☁️ Dockerized, deployed on AWS EC2, tracked with MLflow + DagsHub
- 🛠️ Stack:
Flask·Scikit-learn·Docker·AWS·MongoDB
karamjodh@github:~$ echo "thanks for stopping by"