Rajveer User

I’m an ML and AI systems engineer who focuses on production-grade model workflows: evaluation, latency optimization, and dependable operations. I like turning messy experiments into measurable pipelines with clear instrumentation and regression safety.

At Elevate Labs, I designed PyTorch training and inference pipelines for NLP and computer vision, improving experiment reproducibility with structured preprocessing and automated evaluation. I also optimized inference via latency profiling and batching, cutting average inference time by ~18% across deployed model variants, and built an ML evaluation harness to benchmark and regression-test 5 iterations.

At OutriX, I built an LLM evaluation pipeline processing 1M+ records—automating scoring, regression testing, and failure triage—to reduce experimentation turnaround by 30%. I owned ETL/ELT workflows feeding inference benchmarking dashboards and instrumented end-to-end latency observability with OpenTelemetry, while profiling bottlenecks to reduce p95 latency by ~18%. Earlier, I delivered a distributed anomaly detection pipeline at CDAC India, reducing manual review queue by ~35%.

My projects reflect how I like to build systems around LLMs: AgentOS for agent runtime governance, SemanticMemo for production semantic caching with FAISS-style vector search and verification, and SentinelX for an SLO-aware LLM control plane with policy enforcement and multi-provider failover. I bring a “measure, iterate, ship safely” mindset—supported by recognition like the Amazon ML Summer School ’25 Scholar and CDAC Merit Scholarship.