Built a multi-agent AI system on AWS Bedrock and Strands SDK, orchestrating Claude 3.5 Sonnet to automate configuration drift detection across 1,000+ services, reducing manual reviews by 85% and preventing 12+ production incidents quarterly. Designed 8+ specialized agent tools with Bedrock Converse API for autonomous Git operations and policy validation, achieving 85% reduction in manual reviews. Integrated a precision drift detection engine with line-precise locators and policy-aware validation enforcing 22 compliance rules, achieving 95% accuracy identifying critical deviations and generating actionable remediation for 400K+ configuration files.

AI Engineer at Samvid

July 1, 2024 - August 1, 2025

Engineered a Graph-based Retrieval-Augmented Generation (GraphRAG) system combining OpenAI embeddings with Neo4j knowledge graphs, improving document retrieval accuracy by 60% through integrated semantic and relationship-based search. Crafted and deployed a multi-agent GPT-4 framework for autonomous document analysis, combining specialized agents and a custom orchestration engine to achieve 45% query accuracy and 3× faster knowledge extraction. Formulated a high-performance hybrid search engine by integrating pgvector-based semantic search with PostgreSQL full-text indexing and Redis caching, improving search accuracy by 40% and reducing query latency by 75% across large-scale document corpora. Designed an interactive explainability suite using React Force Graph and custom metadata tracking to visualize reasoning chains, knowledge graph traversal, and source attribution, improving model transparency and enabling stakeholders to trace system outputs.

Data Science Intern at Cisco

January 1, 2024 - June 1, 2024

Constructed and fine-tuned a BERT model to classify unstructured customer return logs into standardized failure categories and applied attention-based visualization for model explainability, reducing resolution time by 25% and cutting operational costs by 15%. Built and deployed an ARIMA-based forecasting model with seasonal decomposition to predict weekly product returns, analyzed 2M+ RMA records to identify return patterns, reduced excess inventory by 20% and improved planning accuracy by 35%. Architected and implemented a serverless ML pipeline on AWS using EC2, Lambda, and S3 to automate end-to-end return log classification, enabling real-time processing of 2M+ customer records monthly and reducing manual intervention by 60%.

Data Science Consultant at Deloitte

June 1, 2022 - July 1, 2023

Led development of a multi-layered LSTM model with attention mechanisms to capture temporal patterns in 500K+ machine vibration sequences, achieving 94% accuracy in predicting equipment failures 48 hours in advance, decreasing unplanned downtime by 20%. Collaborated across three cross-functional teams to implement a real-time vibration monitoring system, enabling automated maintenance scheduling that extended equipment lifespan by 30% and saved $1.2M annually through preventative intervention. Spearheaded model optimization for the LSTM pipeline using quantization, maintaining 92% prediction accuracy and reducing inference time by 65%, enabling deployment on edge devices at equipment sites and provided real-time failure alerts to maintenance teams.

Associate Data Scientist at UnitedHealth Group

July 1, 2019 - May 1, 2022

Devised and revamped over 20 XGBoost-based prediction models to tailor healthcare offers based on customer preferences and interactions, driving a 25% surge in conversion rates and augmenting customer engagement through targeted marketing strategies. Implemented an end-to-end ETL pipeline harnessing PySpark to streamline data processing and elevate operational efficiency, achieving an 80% diminution in Turnaround Time (TAT) while delivering remarkable cost savings of $3M per month. Pioneered adoption of advanced indirect uplift modeling techniques aimed at upgrading campaign effectiveness by cultivating audience segmentation and personalization, contributing to a 35% jump in member engagement and $2M decline in total campaign costs. Managed complete data analysis pipeline using SQL and conducted exploratory data analysis on customer data to identify user behavior patterns, directing to product modifications expanding user experience and customer satisfaction by 15%.