Kerim User

Senior ML Engineer, Team Lead

AstraZeneca

Leading the ML engineering team building AstraZeneca's internal de novo protein design platform, adopted across multiple drug discovery programs. Built scalable pipelines chaining RFdiffusion, ProteinMPNN, and AlphaFold2/Boltz across HPC and cloud GPUs.

ML EngineeringProtein DesignRFDiffusionProteinMPNNAlphaFoldBoltzHPCCloud GPUPipeline DevelopmentIn Silico Filtering

Team Lead, AI Engineering

Leading a 6-engineer AI team automating end-to-end claims processing, reducing manual review time and saving $20M annually. Architected a distributed claims intelligence platform using Airflow, Kafka, and Kubernetes processing 100K+ claims daily.

AI EngineeringClaims AutomationAirflowKafkaKubernetesDistributed SystemsDocument UnderstandingRAGLLMData Processing

Senior ML Engineer

Built LLM-powered extraction pipeline processing millions of academic PDFs, extracting medical taxonomies and entities. Designed RAG architecture achieving 94% accuracy on complex biomedical entity recognition.

LLMRAGDocumentation ExtractionPDF ProcessingNamed Entity RecognitionEmbedding SearchWorkflow Automation

Co-Founder & CEO

Built the entire stack for a secure, self-hosted SaaS MVP in 5 months, including React dashboard, Rust async micro-services, and PyTorch guardrail models. Architected an LLM-agnostic security proxy that cut prompt-injection exploits by 40% for pilot customers.

ReactRustPyTorchKubernetesHelmAmazon EKSMicroservicesLLM SecurityPrompt InjectionSaaSProduct ManagementFundraising

Software Engineer

Implemented real-time flight-control optimizations in C++17 on ARM Cortex, decreasing CPU load 30% and extending UAV endurance by 12%. Integrated computer-vision module (YOLOv7 + CUDA) improving target-tracking accuracy from 82% to 95%.

C++CortexReal Time SystemsFlight ControlComputer VisionYOLOv7CUDAOptimizationUAV

Software Engineer - Researcher

Designed a parallel radar-cross-section simulation on a 1,024-core HPC cluster, reducing computation time from 9 hours to 45 minutes. Optimized C kernels with AVX-512 intrinsics, achieving 3.2 TFLOPS sustained throughput.

HPCParallel ComputingCAVX 512Radar Cross Section AnalysisSimulationOptimization