Hanumanthu Harsha Vardhan

Built a multi-agent system with advanced RAG for industrial machines at a Siemens plant (LangGraph, FastAPI, Pydantic AI, Langfuse), cutting root cause analysis by 73–88% on 16 machines in production. Reached 82.2% cause and 84.3% action alignment on 1,811 historical error codes with an evaluation harness. Retrieval over machine logs and manuals under 200 ms (ChromaDB, on-prem Ollama). Master's thesis project with production deployment.

Raised RAG answer accuracy from 85% to 92% and cut query latency from 500 ms to 150 ms with a local stack (Ollama, FAISS, Sentence Transformers) packaged with Docker and Kubernetes. Scored 0.97 faithfulness on PDF-domain questions with self-designed trust metrics; published at EBL 2026.

Moved solder-joint ok/nok checks for 18 component configurations to continuous inline prediction with a MobileNet, TensorFlow and OpenCV pipeline streaming batches of 512 over ZMQ.

Data engineering, data cleaning and analysis. Automated recurring data workflows with SQL, shell scripts and Python; built BI reports in Python and Tableau; built a Text-to-SQL solution turning natural-language questions into SQL queries.

Coordinated data and systems processes, downtime-reduction work and reporting templates.

Major: AI and Autonomous Driving. Master's thesis submitted October 2026 (multi-agent RAG system for industrial machines, deployed at a Siemens plant).