At Genpact, I architected ICRAS, an 8-agent LangGraph pipeline for contract review and approval. I owned the repository architecture, Pydantic v2 schemas, document ingestion agent, and orchestration and output layer; each run produces evidence-backed audit artifacts and a CLM-ready payload.
I also shipped HelpDesk AI Copilot, a multi-agent RAG assistant that answers routine staff support questions from internal documentation without escalation. It combines LangGraph orchestration with a ChromaDB vector store, a FastAPI backend, and a Streamlit interface.
At Gjirafa Life, I trained and evaluated machine learning models end to end, from dataset preparation and feature engineering through model selection and metric-driven evaluation. I worked with Python, scikit-learn, PyTorch, and TensorFlow.
For my bachelor thesis, I designed and developed a deployed clinic multi-agent system with four specialist agents coordinated through LangGraph. I designed its safety layer so agents propose actions, deterministic Python validates them, and mutating operations require human approval.
