● Developed and deployed an advanced RAG-powered chatbot assistant serving 10,000+ daily queries with 92% accuracy, integrating GPT-4/Claude models with Qdrant vector database for efficient information retrieval. ● Engineered a multi-modal Streamlit application processing 50,000+ monthly user interactions, incorporating vision models (Haiku, Sonnet, Gemini) for visual data extraction and optimized data chunking methods. ● Spearheaded photo moderation workflow optimization, reducing manual review time by 75% through implementation of YOLOv7 for person/children detection and EfficientNet/ViT models for watermark classification. ● Built and deployed automated screenshot/collage classification models achieving 94% accuracy, processing 200,000+ user-uploaded images monthly with real-time auto-cropping capabilities. ● Designed, built, and deployed production-ready AI Agents using Google ADK and LangGraph for internal automation and intelligent task orchestration. ● Implemented multi-agent workflows including tool-execution agent, document understanding agent, and data-retrieval agent; reduced manual operational workload by 60%. ● Architected multi-tool agentic system integrating image preprocessing, agentic RAG (LlamaIndex), and custom tool orchestration; implemented comprehensive observability with LangFuse for end-to-end metric tracking and performance monitoring. Key Projects: 1. Disaster Response Multi-Agent Copilot (Google Agent competition) 2024 ● Developed open-source multi-agent system for automated disaster analysis and response plan generation using sequential agent architecture with specialized task delegation. ● Integrated real-time data sources (news APIs, weather data, geolocation) with LLM-powered analysis agents to generate actionable emergency response strategies. ● Achieved 40% faster response plan generation compared to manual processes; designed for life-critical scenarios requiring rapid decision-making. 2. Multi-Modal RAG Customer Support System 2023-2024 ● Built end-to-end chatbot handling 10,000+ daily queries with 92% resolution rate, integrating text and vision capabilities. 3. Agentic RAG for Internal Process Automation 2024 ● Engineered production-grade agentic RAG system combining custom data ingestion pipeline with LlamaIndex framework, processing 50,000+ internal documents and serving 2,000+ employee queries monthly. ● Built scalable FastAPI async architecture deployed on GCP infrastructure with auto-scaling capabilities.

● Led development of real-time person detection and matting system achieving 30 FPS processing speed, collaborating with 5-member international team across multiple time zones. ● Engineered car detection model for major carwash chain, increasing automated vehicle identification accuracy to 96%. ● Implemented image clustering pipeline for laptop screenshot analysis, processing 1M+ screenshots monthly and computing user KPIs with 89% clustering accuracy. ● Developed and deployed Layout-LM model for NER tasks with 91% F1-score, integrated with Azure cloud services. ● Built medical tumor detection model achieving 94% sensitivity and automated report generation with OCR.

● Led development of predictive models (churn, income, risk, fraud) and real-time scoring systems for 500,000+ customers, improving model accuracy and reducing operational costs while deploying optimized data pipelines across core banking systems.