Machine Learning Engineer

At Mattress Firm, I worked on SleepExpert.AI and ML-driven inventory forecasting platforms, designing and deploying end-to-end machine learning systems spanning data ingestion, feature engineering, model training, evaluation, and production inference across retail and supply chain domains. - Designed and trained recommendation and ranking models (gradient-boosted trees, neural ranking models) to support associate-assisted sales and product guidance, improving engagement and conversion metrics by ~25–30%. - Built NLP pipelines using transformer-based models for semantic understanding, text classification, and document retrieval, enabling fast and accurate access to product, promotion, and operational knowledge. - Developed embedding pipelines (dense vector representations) and similarity search workflows using approximate nearest neighbor (ANN) indexing to power enterprise semantic search. - Contributed to time-series forecasting and demand prediction models leveraging historical sales, seasonality, and store-level signals to improve replenishment and inventory optimization decisions. - Implemented feature engineering pipelines for structured and unstructured data, including normalization, aggregation, windowed features, and categorical encoding. - Built batch and real-time inference pipelines supporting both low-latency online predictions and large-scale offline forecasting jobs. - Implemented model evaluation, monitoring, and retraining workflows to track performance metrics (precision/recall, forecast error) and maintain model accuracy as demand patterns evolved.

At Citi, I developed and productionized ML models for fraud detection, risk scoring, and compliance workflows, with a strong emphasis on feature engineering, model evaluation, and secure ML deployment in regulated environments. - Improved supervised fraud detection models by enhancing feature engineering pipelines, including transaction aggregation, behavioral features, and temporal signals, increasing prediction accuracy by ~25% and reducing false positives by ~30%. - Designed and trained classification and anomaly detection models for transaction risk scoring across highvolume financial datasets. - Built end-to-end ML pipelines covering data ingestion, feature extraction, model training, validation, and inference using Python-based workflows. - Developed NLP-based document classification and summarization models to support compliance review and onboarding automation, reducing manual review time by ~3x. - Implemented model evaluation, monitoring, and retraining strategies to address data drift, concept drift, and changing fraud patterns. - Partnered with risk, compliance, and data teams to ensure ML outputs met regulatory, explainability, and auditability requirements.

At IBM / Kyndryl, I worked on Kyndryl Bridge, builing machine learning models and analytics pipelines for predictive monitoring, anomaly detection, and failure prediction across large-scale enterprise IT systems. - Designed and trained ML models for anomaly detection and predictive failure analysis using time-series telemetry, log data, and performance metrics. - Developed feature extraction pipelines processing tens of millions of events per day, including rolling statistics, seasonality features, and trend-based signals. - Built supervised and unsupervised models to detect abnormal behavior and forecast infrastructure issues before customer impact. - Implemented model evaluation frameworks to validate precision, recall, and false positive rates in production environments. - Integrated ML predictions into automated remediation and decision-support workflows to reduce incident frequency and mean time to resolution. - Collaborated with platform and operations teams to continuously tune models based on real-world feedback and operational data.

At Infosys, this role marked my transition into machine learning engineering, focusing on data preparation, feature engineering, and early ML pipeline development within enterprise environments. - Built data ingestion, cleaning, and transformation pipelines to prepare structured and semi-structured data for ML training and analytics. - Developed feature engineering workflows including aggregation, normalization, encoding, and temporal feature construction. - Supported early supervised learning experiments by improving dataset quality, labeling workflows, and reproducibility. - Implemented batch processing pipelines to generate training datasets and evaluation splits for ML models. - Collaborated with data scientists to operationalize analytics and ML outputs within production enterprise systems.

- Supported data-driven features and backend services for consumer platforms used by millions of customers. - Worked with application telemetry and usage data to improve system reliability and performance. - Collaborated with backend and analytics teams to expose data for reporting and downstream analysis. - Gained early exposure to large-scale data systems and production monitoring.

- Supported the CHI Connect modernization across 100+ hospital locations, improving UI performance by 25% and reducing maintenance overhead by 30%. - Built Angular/JS UI components enabling clinical analytics teams to interact with data systems more efficiently.