Nov 2021 – Present · 4 yrs 7 mos · Noida, Uttar Pradesh, India

• Architected and deployed a Retrieval-Augmented Generation (RAG) system using Sentence-Transformers and LangChain to process structured and unstructured data (Confluence/relational data) into vector embeddings stored in ChromaDB. Delivered real-time infrastructure insights by integrating the RAG system with the OpenAI API, enhancing provides contextually relevant answers by combining retrieved knowledge with large language model capabilities. Developed an intelligent SMS parsing platform leveraging a hybrid ensemble system (FastText, XGBoost, DistilBERT) for real-time and batch classification across 15+ financial domains. Instrumental in creating customer financial profiles, leading to a 28% uplift in targeted outreach for Personal Loan and Bill Reminder teams. Optimized data model performance, achieving 96.5% accuracy through the implementation of advanced ML algorithms and statistical modeling. Architected and deployed a large-scale PYSPARK pipeline processing 10+ Million transactional SMS messages daily across 27+ financial use cases using regex-based Natural Language Processing (NLP). Improved data refresh efficiency by 40% by engineering incremental processing, deduplication, and aggregation jobs (30d/90d/lifetime rolling windows). Automated deployment and cluster management by building a dynamic Airflow DAG factory orchestrating sequential/parallel Spark jobs and managing EMR via Jenkins/Apache Livy API. Ensured PII compliance and platform reliability by implementing Fernet-based encryption for sensitive data and multi-level Slack-based alerting. Developed and deployed an XGBoost gradient boosting model using ensemble methods to predict device failure probability based on CPU, memory, and network metrics. Enabled proactive maintenance through automated anomaly scoring, reducing downtime/issue detection time by 45%. Developed an enterprise-scale Deep Learning PII detection pipeline using Keras neural networks for parallel processing of structured data.

Aug 2020 – Feb 2021 · 6 mos

• Increased user engagement by 60% by deploying a Hybrid User-based Collaborative Filtering Recommendation Engine on Amazon Web Services (AWS) that outperformed existing models in A/B testing. Designed, implemented, and evaluated new models and rapid software prototypes to solve problems in machine learning and systems engineering.