Uchreet insan

Designed and developed a real-time recommendation system leveraging embedding models, hybrid vector search indexing, and AI-driven retrieval techniques to deliver personalized recommendations at scale. Built and optimized batch recommendation pipelines using XGBoost and machine learning algorithms to improve prediction accuracy and customer targeting. Utilized Databricks extensively for end-to-end ML and data engineering workflows, including: Model serving endpoints Unity Catalog management Workflow jobs and orchestration Data pipelines Serverless compute infrastructure ML lifecycle management Used DAB(databricks asset bundles) for the deployment of model, pipelines and workflows. Worked on scalable AI/ML solutions within a financial services environment, applying data science and machine learning techniques to solve business and customer engagement problems. Collaborated with interns and junior team members to automate complex Excel-based workflows and notebooks using Generative AI, agents and models such as OpenAI GPT and Anthropic Claude, improving operational efficiency and reducing manual effort. Contributed to the design and implementation of AI-powered automation solutions by integrating LLM capabilities into enterprise reporting and analytical processes.

-Worked on baggage model for client Pearson airport (GTAA) which predicts if a flight can be delayed because of baggage delay. Was responsible for end-to-end project delivery which included data engineering, data science, and machine learning. -Designed and deployed a complete MLOps stack using Databricks Bundles, automating the deployment of ML models and workflows to UAT and Production environments. -Adopted the Medallion Architecture (Bronze, Silver, Gold layers) to structure data pipelines. -Leveraged Databricks CLI, bundle configuration YAMLs, and CI/CD pipelines to deliver reproducible, environment-aware deployments. -Combined model training, evaluation, registration (via MLflow), and deployment with feature store integration and monitoring for robust end-to-end MLOps workflows. -Created a serving endpoint API to get the results in real time. Developed an AI agent workflow to automatically generate Business Requirement Documents (BRDs) from meeting transcripts and notes, reducing manual effort by 70%. -Integrated NLP techniques and LLMs to extract key action items, stakeholders, and requirements from unstructured meeting data, enhancing documentation accuracy and speed. -Created and deployed multiple time series and regression models like Xgboost , prophet to predict the wait time at different processing areas within the airport and the number of passengers in the flights to departure. -Worked on fine-tuning the machine learning models to get better results using hyperparameter tuning techniques like Grid Search, optuna, and autoML . -Designed workflows and pipelines for automatic training and prediction and generating PowerBI reports. -Trained and deployed a load factor prediction model using azure databricks and MLflow to predict the load factor for the scheduled flights. The model is 89% accurate with MSE (mean squared error) at 5.5 percent.

-Developed an account recommendation model for a financial institution to personalize financial product offerings based on customer behavior, transaction patterns, and demographic insights. Leveraged Databricks Vector Search and embedding-based similarity techniques to generate high-quality, context-aware recommendations by mapping customer profiles and product features into a shared vector space. Integrated the solution into a governed MLOps pipeline using Databricks Bundles, MLflow, Unity Catalog, and Medallion Architecture to ensure secure, reproducible deployment across UAT and Production environments. -Workforce Management & Client Data Transformation: Established pipelines to streamline workforce management for diverse clients. - Developed models for job role classification, role similarity, and ease of role transition. - Utilized AWS to classify job tasks and reduce workforce management costs. - Transformed client data, provided valuable insights using PowerBI, and facilitated competitive analysis. - Led data identification, collection, exploration, and cleaning for model development.

Quality Assurance & Patient Record Enhancement: - Improved the efficiency of Quality Assurance (QA) efforts by creating a system that identifies and rectifies issues in patient records, including outdated and unauthorized pages. - Utilized advanced tools and techniques such as Natural Language Processing (NLP), computer vision, and cloud computing platforms like AWS (EC2, Lambda, ECR) for deploying automated solutions. - Collaborated closely with clients to understand their needs and designed the application's architecture. - Fine-tuned machine learning models through techniques like K-Fold Cross Validation and Grid Search, optimizing their performance. - Conducted data imputation and data extraction, working alongside database engineers to implement ETL processes and writing efficient SQL queries. - Designed an architecture for a cloud-hosted solution on Amazon EC2, enabling streamlined data export to AWS S3. - Created diverse visualizations in Tableau, providing valuable insights through charts and graphs. - Applied advanced Natural Language Processing and Deep Learning algorithms to analyze text data, improving upon existing dictionary-based approaches. - Leveraged various machine learning algorithms and statistical modeling to identify trends and patterns, including decision trees, text analytics, supervised and unsupervised learning, and regression models.

Handwritten Text Digitization & Ticket Classification: - Transformed handwritten medicine names on prescriptions into digital text using PSENET and HTR model. - Employed a Siamese architecture to assess the similarity between different sections of medicine images, achieving a top accuracy rate of 60.3% by utilizing transfer learning with VGG-16. - Digitized handwritten doctor contact numbers through the use of OpenCV and YOLO, achieving an impressive accuracy rate of 95%. - Played a key role in data identification, collection, exploration, and cleaning as part of the model development process. - Conducted experiments with Ensemble methods to enhance model accuracy, utilizing various Bagging and Boosting techniques.

Data Analysis & Store Segmentation: - Detected and extracted text from differently sized boxes on printed forms, employing techniques like dilation, erosion, thresholding, contour analysis, and smoothing. - Developed an application for a retail client to automatically categorize user-raised tickets, classifying them into 20 main categories with approximately 20 subcategories within each. - Conducted analysis of categorical variables, applied statistics and mathematics to optimize performance. - Utilized K-Means and Hierarchical clustering algorithms to segment stores. - Implemented a word cloud generator in Python to analyze and visualize the significance of words in both positive and negative reviews.