Hire Hameed K. - Fine-Tuning Engineer

Building production ML and GenAI systems at CIBC including fraud detection, financial document NLP, RAG pipelines on Azure OpenAI, and agentic AI workflows using LangChain and LangGraph on Azure ML and Databricks infrastructure. · Engineered real-time fraud detection models using XGBoost and LightGBM on Apache Kafka streaming data, improving anomaly precision by 31% under OSFI and FINTRAC regulatory constraints · Built RAG pipelines on Azure OpenAI with LangChain and FAISS vector indexing, enabling document summarization and policy Q&A for 7,500+ internal users with privacy guardrails at the Azure layer · Developed agentic AI workflows using LangGraph with multi-step tool orchestration, memory management, and self-correction loops, reducing compliance review turnaround by 40% · Designed NLP pipelines using Hugging Face Transformers with SFT and LoRA adapters to extract structured entities from mortgage applications and credit memos, eliminating 60% of manual review effort · Implemented MLOps infrastructure on Azure ML and Databricks with MLflow experiment tracking, Delta Lake feature stores, and drift monitoring with automated retraining triggers · Containerized FastAPI inference services with Docker and Kubernetes on Azure with SHAP-based explainability dashboards, supporting responsible AI practices and governance documentation under Canadian banking standards Stack: Python, PyTorch, XGBoost, LightGBM, Hugging Face Transformers, LangChain, LangGraph, FAISS, Azure OpenAI, Azure ML, Databricks, MLflow, Apache Kafka, Delta Lake, FastAPI, Docker, Kubernetes, SQL, Terraform

Applied ML research using PyTorch, TensorFlow, XGBoost, and LightGBM on 500K+ sample datasets, contributing to conference paper submissions and building lab-wide reproducible workflow standards across multiple concurrent research projects. · Built end-to-end preprocessing and feature engineering pipelines on 500K+ sample datasets with variance-based feature selection and cross-validation, reducing model training time by 40% · Applied PCA-based dimensionality reduction, hierarchical clustering, Random Forest, XGBoost, and LightGBM ensemble methods on complex multi-modal datasets, informing 2 conference paper submissions · Trained deep learning classifiers in PyTorch and TensorFlow using CNN and LSTM architectures with ablation experiments tracked in Weights & Biases, achieving 86% accuracy on held-out evaluation sets · Ran hypothesis testing and causal inference using SciPy and StatsModels, producing reproducible workflows adopted as the lab-wide standard across all active projects · Built automated data ingestion and validation pipelines in Python and SQL with GCP BigQuery as the centralized analytical store, recovering 6+ hours of team bandwidth per week Stack: Python, PyTorch, TensorFlow, Scikit-learn, XGBoost, LightGBM, SciPy, StatsModels, Pandas, NumPy, GCP BigQuery, SQL, Weights & Biases, Matplotlib, Seaborn, Git

Delivered end-to-end ML solutions for Fortune 500 clients across healthcare, insurance, and financial services on AWS SageMaker, spanning NLP systems, anomaly detection pipelines, and MLOps infrastructure. · Owned end-to-end ML development across healthcare, insurance, and financial services on AWS SageMaker, building XGBoost and LightGBM pipelines with Bayesian hyperparameter optimization that outperformed legacy rule-based systems by 18% with SHAP-based explainability for audit requirements · Built NLP systems using BERT, spaCy, and Hugging Face Transformers for named entity recognition, semantic classification, and multi-label document tagging, cutting manual annotation effort by 45% across 3 delivery teams · Designed unsupervised anomaly detection systems using Isolation Forest, DBSCAN, and autoencoder architectures on high-volume transactional logs via Apache Spark on AWS Glue with concurrent streaming ingestion · Orchestrated training and deployment workflows with Airflow DAGs and MLflow, implementing drift-triggered retraining with versioning and rollback across production pipelines · Containerized models as Flask REST API microservices on Docker and AWS EC2 handling 10K+ daily inference requests at sub-1.5s latency with GitHub Actions CI/CD Stack: Python, XGBoost, LightGBM, Scikit-learn, Hugging Face Transformers, BERT, spaCy, NLTK, AWS SageMaker, AWS S3, AWS Glue, Apache Spark, Flask, Docker, MLflow, Airflow, PostgreSQL, SQL, GitHub Actions