Phil Piwonka

Recruited to Project Mantle during its early expansion, helping evolve the inference engine that became the backbone of Amazon Bedrock.

• Partner with OpenAI to align Amazon Bedrock's Responses API behavior with OpenAI's first-party implementation.
• Built and launched server-side tool execution through AWS Lambda and Amazon Bedrock AgentCore Gateway connectors, allowing customers to use existing MCP-compatible tools during inference.
• Built and launched Web Search for OpenAI frontier models by integrating Amazon search infrastructure with IAM authorization, usage limits, fairness controls, and production monitoring while keeping customer queries within the AWS environment.

Led the architecture and evolution of Lambda's placement and capacity-management systems as the service expanded across new virtualization platforms, processors, storage configurations, and workload sizes.

• Designed and built Lambda's placement service and engine to manage oversubscription, thermal constraints, and efficient workload placement across the compute fleet.
• Built simulators and utilization models that informed workload entitlements, fleet planning, and placement strategy; partnered with EC2, finance, and capacity-planning teams to improve warm capacity and fleet flexibility.
• Drove Lambda's multi-year migration to Firecracker, rewriting capacity-management and tenancy architecture while the service continued to grow; co-authored the NSDI '20 Firecracker paper.
• Expanded capacity systems across Intel, AMD, and AWS Graviton processors; NVMe- and EBS-backed hosts; and Lambda environments ranging from 1.5 GB to 10 GB.
• Launched Lambda support for container images up to 10 GiB; co-authored the resulting USENIX ATC '23 Best Paper describing a system that has processed hundreds of trillions of invocations for more than one million customers.
• Launched Lambda SnapStart to reduce cold starts for latency-sensitive applications.
• Served as a senior operational reviewer across Lambda, ECS Fargate, and App Runner and joined Lambda's escalation on-call rotation based on broad system and incident-response expertise.

Built customer-facing software and backend services across the original Kindle Fire, cross-platform reading applications, and Goodreads integration.

• Co-developed ACX, an HTML5 framework used to ship Wikipedia Lookup, passage translation, X-Ray, and End Actions across desktop, Android, and iOS Kindle applications.
• Built the original Kindle Fire home-screen launcher and carousel, using custom GPU-accelerated rendering to maintain responsive scrolling across books, music, and video.
• Built the backend integration for Goodreads on Kindle, moving its Ruby on Rails stack into Amazon infrastructure and adding caching and protective services for e-reader and Fire tablet traffic.
• Helped decompose Goodreads' Rails monolith into services and launched Goodreads Deals, a personalized ebook campaign whose batch pipeline was isolated from online customer traffic.

• Developed database providers for Microsoft Sync Framework and integrated them into Visual Studio 2008 and 2010, enabling peer-to-peer synchronization across applications, services, and devices.
• Designed Merge Replication features for SQL Server 2008 and supported enterprise customers through urgent production issues.

• Designed and validated mission-critical numerical and clustering algorithms in a near-real-time, high-performance UNIX parallel-computing environment.
• Improved algorithm performance with multithreading and shared-memory programming using C++, MATLAB, and Perl; led delivery work across multiple customer organizations.