D
Danny Scott
About
Detail
San Antonio, Texas, United States
Résumé
Jobs
verified_user
0% verified
Senior AI/ML EngineerHugging FaceMar 2022 - Current (4 years 6 months)• Led AI product development of an agentic RAG platform using LangGraph, Hugging Face Transformers, and Python, orchestrating AI agents over 20M+ documents and raising answer relevance 34%. • Drove software architecture for AI orchestration frameworks integrating large language models (LLMs) with vector databases, defining Scrum delivery patterns across three product teams. • Fine-tuned open-weight LLMs with LoRA and QLoRA in PyTorch and shipped them through C#/.NET APIs, matching closed-model baselines while cutting per-token inference cost 62%. • Set tech leadership standards for reproducible fine-tuning, prompt versioning, and safe LLM deployment, mentoring 5 ML engineers through the full SDLC. • Optimized AI orchestration frameworks ser- Machine Learning EngineerCognitiveScaleNov 2018 - Feb 2022 (3 years 4 months)• Led AI product development of classification and ranking services in PyTorch and TensorFlow on AWS, exposing predictions through .NET APIs and lifting production F1 from 0.74 to 0.92. • Architected software architecture for ML pipelines with MLflow and Kubeflow on Kubernetes in a Scrum cadence, shrinking iteration time from 2 weeks to 3 days for AI agents. • Built a Kafka feature pipeline feeding an online feature store on Microsoft SQL Server, powering 50M+ daily scoring requests for large language model (LLM) applications. • Deployed real-time inference with BentoML behind .NET APIs, holding p95 latency under 40ms at 15K predictions per second. • Drove tech leadership for model monitoring on Microsoft Azure, reducing silent-failure inci
- Software EngineerCapital OneJul 2015 - Oct 2018 (3 years 4 months)• Built C#/.NET APIs and Python services on AWS powering fraud and credit-decisioning models in a Scrum SDLC, processing 30M+ transactions per day with sub-second scoring. • Shipped the first gradient-boosted fraud model with scikit-learn against Microsoft SQL Server, lifting detection precision 22% and reducing false positives 18%. • Automated model retraining and validation on AWS, cutting release cycles from monthly to weekly and eliminating manual deployment errors. • Developed and maintained internal web applications and dashboards using TypeScript, which supported fraud detection services, enabled real-time monitoring of model performance, and integrated seamlessly with existing C#/.NET APIs and AWS-based backend systems.
Education
verified_user
0% verified
- B.S. in Computer ScienceUniversity of Texas at San AntonioSep 2011 - May 2015 (3 years 9 months)GPA: 3.7 / 4.0
Projects (professional or personal)
verified_user
0% verified
- LAgentic Research AssistantLangGraph, LangChain, Pinecone, PyTorchJan 2024• Designed a supervisor/worker AI agent graph with tool-calling, retrieval, and guardrails that automates multi-step research; adopted by 6+ teams. • Built the RAG layer with fine-tuned embeddings over vector databases and an LLM-as-judge loop, lifting factual-grounding score from 73% to 94%.
- KModel Training & Serving PlatformKubeflow, MLflow, Triton, RayJan 2024• Built a self-service platform for distributed training and one-click deployment on AWS, cutting time-to-production from 3 weeks to 2 days. • Added GPU autoscaling, quantized serving, and canary rollouts, reducing serving cost 40% while sustaining 99.98% availability.