Building production ML systems @ Microsoft AI
Machine learning engineer and applied scientist with 5+ years of experience building scalable, production-grade ML systems across adtech and insurtech with specialization in multimodal modeling, NLP, and ML platform engineering, and deep experience owning solutions from data ingestion to deployment.
At Microsoft, I lead initiatives that improve ad quality and content safety at massive scale, designing models, optimizing pipelines, and modernizing infrastructure with a focus on system simplification, throughput, and reliability. I’ve delivered ML systems that integrate across image, text, and video, working closely with infra, product, and research teams to ship solutions that matter.
Previously at Xandr and Kalepa, I built ML platforms from the ground up, designing model architectures, developing custom data workflows, and deploying inference at scale using tools like Azure ML, Kubernetes, and SageMaker. I’ve worked across CV, NLP, and explainability, and enjoy solving ambiguous problems where ML meets systems engineering.
I thrive in environments where I can own high-impact problems end-to-end, mentor others, and help teams scale fast without compromising quality.
Chat with Clera and we'll introduce you to the right opportunities.
This profile is based on publicly available information. Aakaash is not affiliated with or endorsed by Clera. Privacy policy.

As an Applied Scientist, I lead the design and deployment of large-scale machine learning systems that power ad quality and content compliance at Microsoft. My work spans multimodal modeling, MLOps, and ML infrastructure, with an emphasis on scalability, reliability, and system simplification. - Architected and deployed models across image, text, and video to improve content classification, brand safety, and rejection prediction. - Modernized legacy systems by migrating pipelines to Azure ML and building scalable inference infrastructure. - Led cross-team efforts to improve audit systems, integrate lightweight models (SLMs), and reduce manual review overhead. - Collaborated across product, infra, and modeling teams to deprecate older stacks and standardize new ML workflows. - Trusted to own critical systems end-to-end—known for technical depth, mentorship, and platform leadership. Skills: Applied Machine Learning · Machine Learning · Deep Learning · Microsoft Azure Machine Learning · MLOps · Continuous Integration and Continuous Delivery (CI/CD) · PostgreSQL · Kubernetes · Grafana · Incident Management · Cross-team Collaboration · Technical Documentation · Onboarding · Large Language Model Operations (LLMOps) · Large Language Models (LLM) · Fine Tuning

At Xandr, I built production-ready ML systems to automate and enhance creative ad audits. My work focused on end-to-end ML deployment, data engineering, and model generalization at scale. - Developed transformer-based systems to analyze creatives across image, video, and text. - Built large-scale pipelines to train models on stratified, multimodal datasets. - Integrated embedding stores to support real-time similarity search and zero-shot classification. - Unified fragmented ML workflows into a more maintainable, task-agnostic architecture. - Collaborated with infrastructure teams to deploy and scale ML systems using Azure ML and Kubernetes. Skills: Microsoft Azure Machine Learning · Python (Programming Language) · Deep Learning · Machine Learning · Java · Software Development · REST APIs · Transformer Models · Multimodal · Apache Spark · Kubernetes · Applied Machine Learning

As one of the founding ML engineers at Kalepa, I had the opportunity to shape the core technology stack and build machine learning systems from the ground up. I worked across data ingestion, modeling, explainability, and deployment, helping scale ML capabilities in a fast-moving insurtech environment. - Led technology selection for ML systems, helping establish architectural patterns still in use today. - Built scalable pipelines to ingest and normalize diverse data (social media, news, regulatory filings). - Developed NLP models for extracting policy information, and explainable AI models for underwriter decision support. - Delivered computer vision pipelines for risk detection. - Deployed models on AWS SageMaker and Google Cloud, integrating with real-time risk assessment workflows. - Tackled imbalanced/noisy datasets and engineered solutions for robust training and inference. Skills: Python (Programming Language) · Scikit-Learn · Data Science · MLOps · Jupyter · Machine Learning · Regression Models · Data Pipelines · Google Cloud Platform (GCP) · Amazon Web Services (AWS) · Applied Machine Learning · Applied Technology · Early-Stage Startups

NYU Tandon School of Engineering
- Provided guidance to students toward class lectures, assignments, and projects - Collaborate with the professor on creation and grading of assignments and exams

- Created a Hierarchical CNN model to classify fashion items in images - Performed data engineering techniques to obtain optimal classification result from the Hierarchical CNN model
- Experienced in developing Region-based CNN models for fashion item identification in images - Successfully improved accuracy of previous models by 30% through modifications - Implemented Pose Estimation model to accurately identify human structural poses in images - Developed a testing script to enhance inferencing speed of models by 20% - Created a fallback method to identify fashion apparel in cases of CNN model failure Skills: Python (Programming Language) · Data Science · Deep Learning · PyTorch · TensorFlow · Machine Learning · Convolutional Neural Networks (CNN)

- Provided detailed analysis and actionable feedback for over 4500 projects of students to improve the code and meet standards - Guided and supported students towards completion of Nanodegree in Data Science and Android Development field
- Implemented computer vision algorithms on the Chandrayaan-2 project with Controls Division and Electronics Group team - The algorithm helps the lunar rover identify the distance of objects seen by its stereoscopic camera
Claim it to keep it up to date, or request removal. We're happy to help either way.