Nick McCarthy

About

I am a Senior Specialist Solutions Architect in the Amazon Bedrock team at AWS, currently leading go-to-market, technical enablement, launch readiness, and solution architecture guidance for OpenAI on Bedrock.

Most of my work sits between product, field, and implementation: helping teams understand what to build, helping SAs explain it clearly, and turning model customization patterns into practical guidance customers can use.

I also work closely with the Amazon Bedrock Applied Science team on model customization and reinforcement learning launches, translating new capabilities into field-ready technical guidance and customer patterns.

Before Bedrock, I worked in AWS Professional Services on LLMOps, SageMaker AI platforms, explainability, and applied ML delivery. My background also includes reinforcement learning for finance and UCL degrees in Physics and Machine Learning.

This site collects that trail: AWS technical writing, selected research, education, talks, and project notes with links close to the claims they support.

Timeline

Professional and education milestones, with skills and related artifacts in each dropdown.

ProfessionalEducation

Skills & technologies used

Amazon BedrockAmazon SageMaker AIAmazon Bedrock Custom Model ImportAmazon Bedrock Reinforcement Fine-TuningModel customization workflowsLLMOpsReinforcement learning with verifiable rewards (RLVR)OpenAI-compatible APIs

Bedrock Custom Model Import performance charts

VLM document-to-JSON fine-tuning workflow

Multi-provider GenAI gateway architecture

Currently leading technical GTM from within AWS WWSO for OpenAI on Amazon Bedrock, including launch readiness, SA guidance, and a new OpenAI on Bedrock AWS Technical Field Community upskilling hundreds of technical field members across NAMER and EMEA ahead of GA.
Advised Amazon SageMaker AI's serverless customization launch, contributing RLVR workflow design, HyperPod training recipes, Robin AI preview support, a ~30-person cross-org test program, and a penetration-testing playbook.
Designed a reusable Model Customization Platform Accelerator on Amazon SageMaker AI, publishing AWS Samples code, packaging a Workshop Studio offering, presenting at AWS North America Tech Summit (4.8 CSAT), and delivering a re:Invent Builder's Session with 5/5 CSAT.
Helped build AWS AI League challenges, customer GTM assets, NYC Summit and re:Invent launch workshops, and the Atos customer story; 2025 activations reached 11,326 registrations, roughly 30 lighthouse customers, and 29 influenced opportunities / $16M ARR.
Co-authored enterprise GenAI guidance including mature foundation architecture, the GenAI Lens multi-tenant scenario, model customization readiness frameworks, and two SageMaker AI + Bedrock workshops that drove Talent.com's adoption decision with 4.7/5 CSAT.
Presented a 250-person AWS London Summit breakout, delivered a re:Inforce chalk talk, and led a six-person re:Invent Builder's Session team for an LLMOps platform build-out.
Published technical work on reinforcement fine-tuning, Bedrock Custom Model Import, multi-provider GenAI gateways, and VLM fine-tuning.

Skills & technologies used

LLMOpsAmazon SageMaker AIAmazon BedrockAWS TrainiumAWS InferentiaResponsible AIModel evaluationFine-tuning workflows

Time series forecasting with Amazon SageMaker AutoML

BMW cloud efficiency analytics dashboard

Led delivery of an LLMOps platform for fine-tuning, evaluation, load testing, and deployment of LLMs for Booking.com.
Led development of Booking.com's AI Trip Planner on Amazon SageMaker with the AWS Generative AI Innovation Center, using fine-tuned open-source LLMs for destination recommendations.
Worked with Amazon SageMaker AI, Amazon Bedrock, AWS Trainium, and AWS Inferentia teams on customer feature requests, roadmap feedback, governance, and preview features.
Provided hands-on GenAI advisory through Deep Dive Days and the UKIR GenAI Solutions Launchpad.
Served as sole technical advisor and ML engineer for the winning Aviva GenAI Hackathon team.
Published public work on time series forecasting with Amazon SageMaker AutoML and BMW cloud efficiency analytics with Amazon QuickSight and Amazon Athena.

Skills & technologies used

MLOpsAmazon SageMaker AIAmazon SageMaker ClarifyAmazon SageMaker CanvasExplainable AICI/CD pipelinesGenomics workflowsAWS LambdaAmazon API GatewayAmazon RDSAWS GlueAmazon QuickSight

SageMaker Canvas facies classification workflow

Implemented an MLOps framework for BPX Energy on Amazon SageMaker AI that reduced deployment time and improved model accuracy.
Led an MLOps platform for Ericsson that reduced onboarding time across four ML use cases and avoided significant deployment cost.
Designed a code-promotion CI/CD pattern for AWS partner ML6 that moved ML code between environments rather than promoting opaque model artifacts.
Presented externally at the AWS Manchester User Group and UCISA23, and led EMEA partner ML immersion days.
Influenced 23 AWS Professional Services opportunities worth $31M in bookings through strategic pre-sales discussions with C-level stakeholders.
Designed Amazon SageMaker Clarify experiments for Deutsche Fussball Liga expected-goals models, resulting in an AWS ML Blog post.
Built AstraZeneca deployment and genomics workflows that reduced deployment time and saved scientist hours.
Delivered AI Bench on Amazon SageMaker AI, enabling 300+ researchers to manage over 1,200 ML experiments annually.
Led backend delivery for AWS Dash, an internal KPI tracker using API Gateway, Lambda, Glue, and RDS.
Created and delivered the AI Air Hockey session at the re:Invent 2022 Builder's Fair, attended by 250 participants.

Skills & technologies used

Proximal Policy Optimization (PPO)Deep Deterministic Policy Gradient (DDPG)Soft Actor-Critic (SAC)Portfolio constructionMacroeconomic dataReinforcement learning trading environments

Integrated PPO, DDPG, and SAC into a trading platform that mapped investment signals to target portfolio weights.
Designed RL trading environments and researched neural architectures, action distributions, and macro-data state representations.

Skills & technologies used

Deep reinforcement learningNatural language processingPortfolio optimizationPythonAcademic research writing

Chess pieces from MIT Technology Review SentiMATE article

Thesis: Deep Reinforcement Learning for Portfolio Construction & Optimisation, in collaboration with xAI Asset Management.
Completed the "Deep Learning and Reinforcement Learning" module, taught by Google DeepMind at UCL.
Related public research: SentiMATE, accepted for oral presentation at AIIDE 2019 and covered by MIT Technology Review.

Skills & technologies used

Experimental physicsDoppler-free spectroscopyOpticsUncertainty analysisData analysis

Thesis: Investigation of the hyperfine structures of Rubidium using Doppler-free, frequency-modulated saturated absorption spectroscopy.
Graduated with First Class Honours.

Latest Blogs

Apr 21, 2026

End-to-end lineage with DVC and SageMaker AI MLflow apps

A practical pattern for connecting datasets, training runs, model registry entries, and deployed models into one traceable chain.

Amazon Bedrock RFT best practices diagram

Apr 8, 2026

Reinforcement fine-tuning on Amazon Bedrock: Best practices

Practical guidance for preparing datasets, designing reward functions, monitoring training, and tuning RFT jobs on Amazon Bedrock.

Mar 25, 2026

Reinforcement fine-tuning on Bedrock with OpenAI-compatible APIs

An AWS blog on why OpenAI-compatible APIs matter for model customization workflows on Amazon Bedrock.

All blogs

Speaking

Selected public talks, workshops, and conference sessions across AWS events, customer forums, and research venues.

NYC Summit and AWS re:Invent / 2025 / New York / Las Vegas

AWS AI League launch workshops

Helped build hands-on model-customization challenges, customer GTM assets, service-team feedback loops, and launch workshops for AWS AI League.

11,326 registrations, roughly 30 lighthouse enterprise customers, 29 influenced opportunities / $16M ARR.

AWS re:Invent AIM319 LLMOps workshop graphic

AWS re:Invent / 2025 / Las Vegas, USA

LLMOps and SageMaker AI fine-tuning workshop

Built and delivered a hands-on LLMOps workshop around automated fine-tuning, evaluation, and deployment workflows on Amazon SageMaker AI.

50 builders, 5 instructors, and 150 GPUs across the workshop environment.

AWS re:Inforce / 2025 / Philadelphia, USA

Responsible AI and secure GenAI agents

Delivered security-focused GenAI sessions covering responsible AI patterns and practical controls for agentic systems.

Speaker CSAT 4.84.

AWS London Summit HyperPod breakout session

AWS London Summit breakout session / 2025 / London, UK

Accelerate AI/ML workloads with Amazon SageMaker HyperPod

Co-presented a breakout session at the AWS London Summit on monitoring, prioritizing, and governing AI development tasks with Amazon SageMaker HyperPod.

AWS Summit technical session.

Education

UCL degrees, thesis work, and related research links.

Sept 2018 - Aug 2019

Machine Learning MSc

University College LondonDistinction (77%)

Focused on reinforcement learning, statistical learning, and practical ML systems. Related public research includes SentiMATE, accepted for oral presentation at AIIDE 2019.

Thesis:Deep Reinforcement Learning for Portfolio Construction & Optimisation, in collaboration with xAI Asset Management.

SentiMATE arXiv MIT Tech Review

Sept 2014 - Jun 2017

Physics BSc

University College LondonFirst Class Honours

Built a quantitative grounding in experimental physics, optics, spectroscopy, and data analysis.

Thesis:Investigation of the hyperfine structures of Rubidium using Doppler-free, frequency-modulated saturated absorption spectroscopy.

Papers

Publication-style research work linked back to the relevant source.

Amazon Science paper / 2021

Amazon SageMaker Clarify: Machine Learning Bias Detection and Explainability in the Cloud

Amazon Science publication on bias detection and explainability in Amazon SageMaker Clarify.

AIIDE publication / 2019

SentiMATE: Learning to play Chess through Natural Language Processing

Research accepted for oral presentation at AIIDE 2019 on learning chess behavior from natural language.

Let's Connect

Open to collaborations, project conversations, and thoughtful feedback loops.

Email LinkedIn GitHub