Kartik Sarda

Architected and deployed an intelligent web automation system integrating Playwright MCP server with Celery worker using stdio transport, enabling autonomous agents to navigate websites, download CSV files, and process data for SQL query generation via Vanna AI connector; Designed and implemented Kubernetes sidecar architecture to overcome K8s stdio limitations, deploying Playwright MCP server alongside Celery worker containers with streamable HTTP transport (localhost:8931/mcp), ensuring seamless inter-container communication in production environment; Enhanced Vanna AI model accuracy by 28% by diagnosing and fixing embedding dimension mismatches, reducing hallucination rate from 15% to 4% in SQL query generation through optimized RAG pipeline configuration; Integrated MySQL source connector to establish seamless connectivity between the company's Amplifi application and database, enabling effective data flow for model training and real-time analytics; Developed agentic tools leveraging PaLM and ReAct frameworks to enable advanced reasoning and decision-making within AI workflows, improving autonomous task completion rate by 40%; Integrated all components using FastAPI and Docker,

Completed comprehensive training in NLP and Generative AI, covering transformer architectures, attention mechanisms, RAG systems, and LLM fine-tuning techniques; Built hands-on projects including PDF Q&A systems (LangChain + OpenAI), web scraping agents (CrewAI), and fine-tuned speech recognition models to apply theoretical concepts in production-ready applications; Implemented model evaluation pipelines, benchmarked custom-trained NLP models against baseline architectures, and documented experimental results to inform future iterations at ThoughtsWin