Ramjan Khandelwal

I’m an AI Research Associate at Stryker, building systems that move from research to production with tight latency, reliability, and evaluation discipline. I’m also actively developing agentic RAG and ML tooling through hands-on projects while completing my B.Tech in Computer Science.

At Stryker, I built a real-time inference engine deployed on production hospital beds using multiprocessing scale-out with latency monitoring, extending it to multi-bed rollout while sharing one core across live and replay paths. I designed a 2-level stacking ensemble for posture classification, achieving 92.5% accuracy with leave-one-patient-out validation, and improved out-of-distribution hardest-class accuracy by +15.6 pp using real-time personalized normalization (streaming statistics) at inference without retraining.

I focus on the engineering details that prevent silent failures—hardening the pipeline against train/serve skew and label leakage, and verifying deploy-vs-offline agreement at 99.8% across beds. I’ve also shipped robust MLOps workflows on AzureML (versioned Data Assets and MLflow Model Registry), created an agentic RAG evaluation harness with LangGraph (catching planner regression that faithfulness missed), built an agentic image editing environment, and even implemented a Redis-compatible in-memory store to sharpen systems performance instincts.