Shawal Khan
AI/ML Engineer | Web Developer
AI/ML engineer skilled in Python, scikit-learn, and full-stack web dev. Experienced in end-to-end ML pipelines and seeking entry-level AI/ML roles.
About
Aspiring AI and ML Engineer with hands on experience building end to end machine learning pipelines, deploying models, and applying regularization and class balancing techniques to biomedical and financial datasets. Seeking entry level opportunities in AI and machine learning engineering roles.
Skills
Languages: English, Urdu, Pashto
Education
BS Computer Science
Institute of Management Sciences (IMSciences), Peshawar
Experience
Machine Learning Engineer Intern · CodeClub Software Solutions
May 2025 – June 2025Designed and implemented end-to-end ML pipelines covering data preprocessing, feature engineering, model training, and evaluation using Python, scikit-learn, and pandas. Built and deployed a loan default prediction model with a Streamlit front-end, achieving strong cross-validation performance through regularization tuning and SMOTE-based class balancing. Explored SVM and KNN classification, conducted hyperparameter sweeps, and produced IEEE-style technical reports; applied EEG-based seizure prediction using deep learning on high-dimensional imbalanced biomedical data.
Projects
LLaMA-XR Extended - Radiology Report Generation (LLM Fine-tuning)
Reproduced a 2026 arXiv paper (no code released) fine-tuning LLaMA 3.1 8B with QLoRA for chest X-ray report generation; found the paper's stated learning rate failed to reproduce, diagnosed the cause, and validated a fix across a 590-example test set. Built two evaluation methods missing from the original paper: a clinical-accuracy F1 scorer and a hallucination-detection check; used them to quantify a model grounding failure (clinical F1 0.018) and motivate a proposed architectural fix. Stack: Python, PyTorch, Unsloth, QLoRA, Hugging Face Transformers/TRL, DenseNet-121, Kaggle/Colab.
View project →EEG Seizure Prediction - Regularisation & Generalisation Study
Investigated preprocessing strategies, model complexity, and regularisation (L1, L2, Elastic Net) across 3 EEG datasets (5–20% seizure rate); designed two distinct preprocessing pipelines and benchmarked generalisation under class-imbalanced conditions using SMOTE. Stack: Python, scikit-learn, pandas, NumPy, matplotlib, imbalanced-learn.
View project →Loan Default Prediction - ML Pipeline
Built a complete classification pipeline (EDA, preprocessing, feature engineering) with SMOTE for class imbalance; compared Logistic Regression, SVM, and KNN with regularization sweeps and cross-validation metrics; deployed via Streamlit on GitHub.
View project →We use cookies to improve your experience and show relevant ads. By continuing, you agree to our use of cookies.Cookie Policy