Senior AI Engineer at Insight Enterprises

Architected and led a multi-agent AI analytics platform on Google Vertex AI using Google Agent Development Kit (ADK) and AG-UI, delivering natural-language business intelligence, reducing time-to-insight by 60% and increasing industry penetration by 20%. Built a high-performance text-to-SQL agent using Claude and Snowflake Semantic Views, reducing query latency and compute costs while improving SQL accuracy. Designed hybrid Graph-RAG agents with Neo4j and Vertex AI Search for grounded retrieval across structured/unstructured enterprise knowledge with improved relevance and explainability. Implemented governed Snowflake pipelines and an MCP gateway via Apigee API Management and Vertex Agent Engine to expose secure, multi-tenant business data as reusable AI tools across 30+ customer environments with centralized governance. Productionized on Google Cloud with Docker, GKS, Terraform/Terragrunt, GitHub Actions, and Vertex AI Safety Controls, unifying runtimes and improving security against

Senior AI/ML Engineer at Tutor Me Education

Architected a production-grade LLM-powered multi-agent conversational AI voice tutor using LiveKit for real-time sessions. Integrated ASR (Deepgram), TTS (ElevenLabs), LangChain, and LangGraph to deliver personalized tutoring for hundreds of students, improving session completion rates. Increased engagement and tutor scalability by 60% through GPT-4o fine-tuning on 100K+ conversation scripts, plus RAG pipeline optimization and agent orchestration. Built domain-specific RAG using PgVector, sentence-transformer embeddings, and HNSW indexing to improve factual accuracy by 40% across 10M+ monthly queries. Implemented multi-agent orchestration with LangChain/LangGraph and ReAct/CoT prompting to reduce routing latency by 50% and improve adaptive tutoring decisions. Designed large-scale Databricks data pipelines processing 10M+ daily tutoring interactions to accelerate analytics and training dataset generation. Implemented multimodal pipelines combining OCR, object detection (YOLOv8), OpenCV,

Python ML Engineer at Stripe

Built and scaled ML-based fraud detection systems powering Stripe Radar using gradient boosting and deep neural networks (Python, TensorFlow, XGBoost), reducing fraudulent transactions by ~30% while maintaining authorization rates. Architected real-time risk scoring infrastructure using Apache Spark and Kafka for sub-100ms inference at high daily transaction volume. Developed sequence-based deep learning models (LSTM/temporal models) to detect evolving fraud patterns, increasing early fraud recall by ~25% in high-risk segments. Built NLP-based classifiers to automate dispute and chargeback workflows, reducing manual review workload by 40%+ and improving resolution speed. Developed scalable MLOps pipelines using Docker, Kubernetes, and AWS for automated deployment, monitoring, and rollback in high-throughput production. Partnered with data engineering to build ETL pipelines and feature stores, reducing data latency by ~40% and improving downstream model performance.