Sankalp Mittal

I’m a Software Engineer III in Cloud AI/ML at Google India, where I own end-to-end model building for Nightingale—a production Agentic RAG pipeline supporting Google Cloud SaaS support. I’ve directly connected theoretical RAG and graph retrieval research to applied production systems, including a reported 35% reduction in on-call triage time.

I build multimodal and agentic capabilities using ReAct-pattern loops, self-reflection, and VLM capabilities, and I focus on efficient inference through model distillation and quantization (TensorRT-LLM, vLLM). I’ve also designed LLM evaluation systems—automated evals, shadow testing, human feedback, and regression—and I craft prompt stacks (CoT/ToT) to generate signals for NER, sentiment analysis, and text classification.

I develop reliability and operations into the core of the systems: I built an RCA agent for root-cause analysis, mitigation steps, and failure paths, plus full LLMOps observability using Cloud Monitoring, Cloud Trace, and Grafana with drift detection. I led ML lifecycle automation on GCP (Vertex AI Pipelines, Cloud Composer, GKE) using Terraform/IaC, Docker/Kubernetes, and I defined SLO/SLIs and led incident response.

Before Google, I was a Research Engineer at IIT Hyderabad (ZF Technologies), where I worked across strong research-to-production themes: Text2Arch (structured visual diagram generation with multimodal VLMs and graph parsing/eval metrics), LENS (locality-sensitive attribution robustness and interpretability for safety-critical vision), and iGuard/AdaFlow (knowledge distillation and anomaly detection deployed in a memory-constrained P4/Intel Tofino data plane at 6.4 Tbps and 532ns latency). Earlier at Ericsson India, I built end-to-end 5G traffic forecasting and real-time streaming inference stacks, using distributed feature engineering and MLOps observability to cut deployment cycles by 40%.