Mayank vanik

Developed a scalable AI-powered service that generates complete, runnable projects from user prompts using agentic AI, multi-LLM coordination, and MCP-based automated code execution and version control. Designed and deployed a robust, scalable AI-driven service leveraging agentic AI architectures (LangGraph) to build multiple intelligent tools and agents. The system dynamically generates runnable code in response to user queries, automatically version-controls it via GitHub integration through Model Context Protocol (MCP) for seamless push operations and change tracking. Using MCP, generated code can also be executed on virtual cloud environments instantly, enabling rapid testing and deployment without manual setup. The platform integrates MongoDB, Vector Databases, and Graph Databases for diverse workloads, supporting advanced retrieval, storage, and relational data processing. The entire infrastructure is deployed and managed on AWS, with a dedicated AI Agent Server orchestrating operations, ensuring high availability, scalability, and resilience. Multiple Large Language Models (LLMs) are employed in a coordinated agent framework, each specialized in distinct capabilities to maximize accuracy, efficiency, and contextual reasoning. Built GPT-4 powered chatbot using LangGraph + LangChain, reducing sales workload 75% and boosting conversions by 45%. Integrated real-time RAG with Weaviate + Tavily; achieved significant accuracy and reduce-2s response time. Deployed on AWS with Docker + CI/CD, optimized for UAE with multilingual, localized features.

Developed RASA-based HR chatbot with custom spaCy NLU, achieving 95% intent accuracy across 200+ employees. Automated WFH/leave workflows with LLM-driven form filling and API integrations, reducing processing time by 70%.

Built YOLOv8-based CV pipeline for face detection and clustering in 2K+ marriage album images with 88% accuracy. Applied geometric filtering and FaceNet embeddings for front-face selection and high-precision (91%) matching. Deployed real-time system on AWS with OpenCV and autoscaling support for efficient large-scale processing.

Developed OCR + GPT-4 based system for solving STEM MCQs with good accuracy and 7s avg response time. Used Tesseract OCR with preprocessing for text extraction accuracy across varied question formats. Deployed Streamlit web app with user login, answer explanations, and progress tracking features.