Fintech / Credit risk 2026
A causally-audited reinforcement-learning pipeline for loan/insurance offer optimization, built on real BPI Challenge 2017 event data.
Offers in demand model 42,821
Events / applications 193,849 / 31,509 PM4py EconML Apache Beam Kubernetes Kubeflow Process mining Causal inference View case study Privacy-preserving ML / Edge 2025
Next-word prediction trained across distributed clients where only differentially-private model updates ever leave the device.
Privacy budget after 15 rounds ε ≈ 1.12
Top-5 next-word accuracy 19.0% Flower Opacus Kafka Flink Scala Redis Grafana View case study AI platform / Infrastructure 2026
Qwen2.5-3B-Instruct benchmarked end to end , distributed DPO training, hand-written CUDA kernels, a three-way inference-engine comparison, Triton serving and Kubernetes autoscaling.
Fused CUDA kernel speedup 2.43x
Throughput gap vs. ONNX Runtime GenAI 7.2x DPO RLHF LoRA Ray Horovod CUDA C++ ONNX Runtime View case study