At Yukosa, I built MetaFlow’s MCP integration system, including OAuth 2.0 connection flows and support for remote MCP servers. I also audited and hardened enterprise connectors through live testing against real infrastructure.
For MetaFlow’s AI Assistant, I built a LangGraph agent that turns natural-language requests into executable workflows. On-demand retrieval cut LLM operating costs by about 46%, and I made conversation sessions durable across restarts and server instances.
At The Strelema, I designed and deployed a production RAG system using FAISS and Qdrant, improving retrieval precision by 40–55% and reducing latency to 300 ms. I also automated document ingestion on AWS and built a multilingual sentiment-analysis pipeline; earlier, as a Machine Learning Research Intern at College of Engineering Pune (COEP), I contributed to audio classification and source-separation research.
