Nikhil Kumar

I’m a Senior Data Engineer and founding team builder at Khyaal, where I architect PB-scale, unified event-driven CDP and AI agent ecosystems for 1M+ users. I focus on reliability, measurable business impact, and production-grade systems that can recover cleanly and scale.

At Khyaal, I built a metadata-driven data platform using Airflow, PySpark, and BigQuery—replacing Mixpanel and saving INR 3.5M ($42K) annually. I also engineered fault-tolerant Kafka consumers for 10M+ daily events with automated schema evolution and field collision detection, achieving 67 production schema breaks prevented with <10s recovery. On the infrastructure side, I deployed multi-cloud Kubernetes across AWS EKS and GCP GKE via Terraform, delivering 3-AZ high availability and 98% platform uptime.

I lead real-time streaming and agent orchestration end-to-end. I designed a PyFlink + Kafka journey orchestration engine for exactly-once semantics at <80ms p99 latency, improved user activation by 35%, and supported 3M+ journey executions per month. I also architected distributed agent orchestration (Airflow + Kafka + ephemeral Kubernetes pods), built production RAG components using LangGraph and Qdrant stacks, and shipped a real-time AI Voice sales engine using the Gemini Live API with bidirectional streaming for 5+ Indic languages (78% accuracy improvement and <80ms p95 latency).

Before Khyaal, I worked at MadStreetDen as a Machine Learning Engineer, delivering enterprise ML for high-valuation financial and logistics clients—building an OTS (One-Time Settlement) engine that drove a 30% YoY collection increase and $10M revenue uplift, and reducing fraud claims by 42% for a $250M+ client. Earlier, I fine-tuned LLMs with LoRA in a research internship (40% accuracy improvement and 80% precision) and improved ecommerce product categorization to 94% accuracy, cutting manual review effort by 24%. I’m proudest of the teams I’ve helped build and the engineering standards I’ve established—code reviews, CI/CD, documentation, and incident playbooks—because great systems are engineered, not wished into existence.