Ajay reddy

Architected an enterprise-grade clinical intelligence platform built on Retrieval-Augmented Generation combining Azure OpenAI models, Hugging Face transformers, and vector embedding pipelines to retrieve insights from structured claims data and unstructured clinical text Designed distributed ETL pipelines using Databricks, PySpark, and Apache Airflow processing PHI-compliant datasets at scale, improving processing throughput by 35% and enabling near-real-time analytics Fine-tuned domain-adapted transformer architectures for clinical NLP tasks including named entity recognition, ICD code extraction, and risk stratification, increasing prediction accuracy by 18% Engineered semantic search architecture using embedding models and vector databases that reduced clinical query response time by 40% while simultaneously lowering token usage and inference costs Developed predictive analytics models for readmission risk scoring, utilization forecasting, and cost prediction using neural networks, gradient boosting, and clustering algorithms Built production-grade APIs and microservices using FastAPI and Node.js enabling secure access to models across enterprise applications and clinical dashboards Deployed containerized AI services via Docker and Kubernetes (AKS) integrated with automated CI/CD pipelines using GitHub Actions and Azure DevOps for version-controlled model rollouts Implemented experiment tracking, model registry integration, validation workflows, and reproducibility pipelines using MLflow and Azure Model Registry Applied dimensionality reduction techniques including PCA, UMAP, and t-SNE to optimize high-dimensional clinical datasets and improve downstream model performance Collaborated with clinicians, product managers, and compliance teams to translate domain requirements into scalable AI solutions aligned with HIPAA and governance frameworks Mentored 3 junior ML engineers on LLM optimization, model deployment, and MLOps best practices

Led migration of legacy SAS-based credit risk models into optimized Python and PySQL pipelines, maintaining 100% logic fidelity while improving execution speed by 300% Architected scalable data engineering pipelines capable of processing 50+ TB of structured transactions and unstructured financial documents using Databricks and Azure Data Lake Built hybrid machine learning risk scoring system combining XGBoost and deep neural networks that improved default prediction performance (AUC) by 18% compared to logistic regression baseline Developed enterprise RAG system using Azure OpenAI, LangChain, and semantic search infrastructure that reduced inference latency by 40% and decreased token cost by 30% through prompt compression and caching strategies Implemented advanced prompt engineering frameworks and LangSmith tracing workflows to evaluate LLM outputs and ensure financial accuracy while minimizing hallucinations Designed fairness and bias detection pipelines using Fairlearn ensuring all AI-driven credit decisions adhered to regulatory lending standards Managed full machine learning lifecycle including experiment tracking, testing, versioning, and deployment via Azure DevOps + MLflow Automated financial reporting workflows that reduced manual audit effort by 40%, saving millions annually in operational cost Developed interactive Power BI dashboards connected to Azure SQL enabling real-time visualization of portfolio risk indicators for stakeholders and audit partners Communicated complex technical findings to executive-level stakeholders through data storytelling and analytical presentations

Partnered with marketing and sales stakeholders to identify drivers of customer churn through large-scale exploratory data analysis and feature engineering Extracted, cleaned, and transformed CRM datasets using SQL and Python for downstream modeling workflows Built classification models including logistic regression, random forest, and gradient boosting to predict churn risk across customer segments Achieved 15% improvement in prediction accuracy compared to baseline models through hyperparameter tuning and feature optimization Deployed production machine learning solution using Docker integrated into CRM systems for automated prediction delivery Developed analytics dashboards visualizing churn predictions and key business drivers enabling data-driven decision-making across teams

● Using Python, Jupyter Notebook, TensorFlow, PASCAL VOC, and Selenium, I was able to provide significant outcomes in object detection and image processing during my internship. I demonstrated my adeptness in web scraping and data acquisition by skillfully extracting large-scale image data from several online sources using the Selenium web driver. I showed a resolute dedication to accuracy in item tagging by carefully using the PASCAL VOC annotation standard. I demonstrated my proficiency in data preprocessing and quality control by ensuring annotation uniformity using tools like labelImg and VGG Image Annotator. ● My expertise included sophisticated data pretreatment methods and a thorough comprehension of the YOLOv4 object detection concept.where I achieved excellent accuracy by using transfer learning to fine-tune the model. By assessing model performance using measures such as mean average precision, I demonstrated my adeptness in analysis. I contributed to real-world deployment by showcasing my expertise in testing and evaluating models. I stayed up to date on AI breakthroughs since I'm committed to ongoing improvement, which improved workflow efficiency. I successfully shared my work and insights in a cooperative team setting, which helped the project succeed.