Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

I am Arash Shahmansoori, a senior AI/ML research engineer and technical lead with over 10 years in machine learning and statistical signal processing, and more than 5 years in deep learning, LLMs, and generative AI. I design, train, and ship agentic LLM systems and multi-GPU/distributed pipelines end-to-end—from data strategy and modeling to deployment, monitoring, and cost/performance optimization—in cybersecurity and voice assistant products. I bring hands-on MLOps/LLMOps across AWS/Azure/GCP/Kubernetes; I'm expert with PyTorch/Transformers/DeepSpeed, Ray/Kubeflow/MLflow, Triton/ONNX/TensorRT, and modern RAG/vector stacks. I'm known for turning research into reliable, scalable production: improving accuracy, lowering latency and unit cost via retrieval, distillation, and quantization, and hardening safety/privacy with automated evals, red-teaming, guardrails, and differential privacy/federated learning. I enjoy mentoring, setting standards, and partnering with product/security/legal to ship compliant, trustworthy AI.…I am Arash Shahmansoori, a senior AI/ML research engineer and technical lead with over 10 years in machine learning and statistical signal processing, and more than 5 years in deep learning, LLMs, and generative AI. I design, train, and ship agentic LLM systems and multi-GPU/distributed pipelines end-to-end—from data strategy and modeling to deployment, monitoring, and cost/performance optimization—in cybersecurity and voice assistant products. I bring hands-on MLOps/LLMOps across AWS/Azure/GCP/Kubernetes; I'm expert with PyTorch/Transformers/DeepSpeed, Ray/Kubeflow/MLflow, Triton/ONNX/TensorRT, and modern RAG/vector stacks. I'm known for turning research into reliable, scalable production: improving accuracy, lowering latency and unit cost via retrieval, distillation, and quantization, and hardening safety/privacy with automated evals, red-teaming, guardrails, and differential privacy/federated learning. I enjoy mentoring, setting standards, and partnering with product/security/legal to ship compliant, trustworthy AI.

Arash Shahmansoori

AI Engineer, Data Scientist, Full Stack Developer, +2





I am Arash Shahmansoori, a senior AI/ML research engineer and technical lead with over 10 years in machine learning and statistical signal processing, and more than 5 years in deep learning, LLMs, and generative AI. I design, train, and ship agentic LLM systems and multi-GPU/distributed pipelines end-to-end—from data strategy and modeling to deployment, monitoring, and cost/performance optimization—in cybersecurity and voice assistant products. I bring hands-on MLOps/LLMOps across AWS/Azure/GCP/Kubernetes; I'm expert with PyTorch/Transformers/DeepSpeed, Ray/Kubeflow/MLflow, Triton/ONNX/TensorRT, and modern RAG/vector stacks. I'm known for turning research into reliable, scalable production: improving accuracy, lowering latency and unit cost via retrieval, distillation, and quantization, and hardening safety/privacy with automated evals, red-teaming, guardrails, and differential privacy/federated learning. I enjoy mentoring, setting standards, and partnering with product/security/legal to ship compliant, trustworthy AI.…I am Arash Shahmansoori, a senior AI/ML research engineer and technical lead with over 10 years in machine learning and statistical signal processing, and more than 5 years in deep learning, LLMs, and generative AI. I design, train, and ship agentic LLM systems and multi-GPU/distributed pipelines end-to-end—from data strategy and modeling to deployment, monitoring, and cost/performance optimization—in cybersecurity and voice assistant products. I bring hands-on MLOps/LLMOps across AWS/Azure/GCP/Kubernetes; I'm expert with PyTorch/Transformers/DeepSpeed, Ray/Kubeflow/MLflow, Triton/ONNX/TensorRT, and modern RAG/vector stacks. I'm known for turning research into reliable, scalable production: improving accuracy, lowering latency and unit cost via retrieval, distillation, and quantization, and hardening safety/privacy with automated evals, red-teaming, guardrails, and differential privacy/federated learning. I enjoy mentoring, setting standards, and partnering with product/security/legal to ship compliant, trustworthy AI.

Available to hire

I am Arash Shahmansoori, a senior AI/ML research engineer and technical lead with over 10 years in machine learning and statistical signal processing, and more than 5 years in deep learning, LLMs, and generative AI. I design, train, and ship agentic LLM systems and multi-GPU/distributed pipelines end-to-end—from data strategy and modeling to deployment, monitoring, and cost/performance optimization—in cybersecurity and voice assistant products.

I bring hands-on MLOps/LLMOps across AWS/Azure/GCP/Kubernetes; I’m expert with PyTorch/Transformers/DeepSpeed, Ray/Kubeflow/MLflow, Triton/ONNX/TensorRT, and modern RAG/vector stacks. I’m known for turning research into reliable, scalable production: improving accuracy, lowering latency and unit cost via retrieval, distillation, and quantization, and hardening safety/privacy with automated evals, red-teaming, guardrails, and differential privacy/federated learning. I enjoy mentoring, setting standards, and partnering with product/security/legal to ship compliant, trustworthy AI.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Language

English

Fluent

Work Experience

Independent Project Researcher at Independent Projects

September 1, 2025 - Present

Leading and contributing to agentic AI systems and related research across multiple domains; ongoing development and evaluation with emphasis on safety, privacy, and tool-use with evolving architectures (see LinkedIn for full detailing).

Lead Machine Learning Research Engineer at ProofPoint Ireland Limited

April 1, 2025 - September 1, 2025

Proposed and led a graph-based data framework to generate high-fidelity training data for LLMs focused on phishing detection and cybersecurity threat understanding. Architected a unified multi-GPU distributed training framework for LLMs on large-scale email corpora; productionized workflows with Docker/Kubernetes, GitOps/ArgoCD, and CI/CD. Built robust data pipelines across AWS (S3, Athena, Glue) to support scalable training/evaluation cycles for threat reasoning use cases. Mentored junior engineers and improved SDLC via design reviews and pairing.

Lead Machine Learning Research Engineer at Tata Consultancy Services Ireland Limited

July 1, 2024 - April 1, 2025

Proposed and led development of an agent-based GenAI system capable of task decomposition, tool-use, and self-correction to fulfill complex information needs. Designed internal state/memory strategies to reduce generation cost and latency while improving answer quality and consistency. Delivered end-to-end solution on AWS with production-grade MLOps: data pipelines, deployment, monitoring, and CI/CD. Mentored junior engineers and provided technical leadership in architecture, experimentation, and delivery best practices.

Senior Post-Doctoral Machine Learning Research Engineer at University College Cork

November 1, 2020 - June 1, 2024

Developed a dynamic consent management system for voice assistant platforms using deep learning, enabling privacy-preserving training and adaptive consent handling. Deployed privacy-aware training approaches and ML/LLM solutions on AWS with SageMaker, MLOps/LLMOps practices, and Hugging Face integrations. Designed cognitive multi-agent systems for autonomous problem solving and planning; conducted research on hierarchical agents for multi-modal synthesis, analysis, and retrieval; contributed open research on LLM planning and multi-modal task decomposition.

Machine Learning Research Engineer at Mitsubishi Electric R & D Center Rennes

January 1, 2019 - December 1, 2019

Built machine learning methods for black-box channel learning, sparse Bayesian learning for massive MIMO, and probabilistic amplitude shaping.

Post-Doctoral Research Engineer at University of Rennes 1

July 1, 2017 - December 1, 2018

Developed algorithms for localization and tracking using millimeter-wave lens MIMO in 5G systems and participated in mmWave massive MIMO measurement campaigns.