Connor
Gramazio

About

I am a senior software engineer (SDE3) and a technical lead for Audible's text-to-speech (TTS) and machine translation (MT) organization. My work spans machine learning engineering, distributed services, and front-end development. Since joining Amazon in 2017, I have built language technology for Audible, Amazon Pharmacy, and Alexa:

• At Audible, I lead TTS and MT initiatives that directly serve CEO Bob Carrigan's vision: “every book is an audiobook in every language”. I designed, built, and launched Audible's machine translation system from scratch. My work in TTS and MT spans subjects like computational linguistics, agentic architectures and LLM judges, inference and prompt optimization, and customer-facing full stack development. I also collaborate across roles and organizations to bring state-of-the-art R&D into production quickly, such as ChainAlign, a cross-lingual aligner.
• At Amazon Pharmacy, I designed and standardized machine learning architecture patterns across the Pharmacy Operations technology organization. These patterns laid the groundwork for an order-of-magnitude reduction in error rates (see Nature Medicine) and for higher operational throughput.
• At Alexa AI (now AGI), I led the C++ and Python development that let BERT models run inference on CPUs in under 5 ms and train across distributed GPUs, powering the inner-most-loop of Alexa's natural language understanding (NLU).

I am most at home in organizations that challenge me to learn new skills, let me bridge roles and disciplines, and take meaningful action to improve diversity, equity, and inclusion.

Before moving into industry, I completed a Computer Science PhD at Brown University as a National Science Foundation Graduate Research Fellow. My dissertation focused on visualization design and drew on a range of research methods, including systems building, algorithm design, and human-subjects experiments.

Contact

You can get in touch through LinkedIn. I also correspond through personal e-mail, but I do not list it publicly given widespread scraping+spam.

If you recruit for blockchain, “web3,” or cryptocurrency roles, please do not contact me. Web3 is going just great.

Research

While in academia, I focused on information visualization design techniques and theory. My research produced eleven publications and several conference posters. For a publication list, please see my Semantic Scholar, Google Scholar, or dblp profiles.

Much of my dissertation focused on color's role in effective visualization. My talk at OpenVisConf 2017, Empowering Effective Visualization (Color) Design, highlights:

• Colorgorical: a tool that automatically generates categorical color palettes, balancing legibility and aesthetic preference. (use) (paper) (source)
• Chromaticity: a tool for building color palettes by hand, with a color picker that spans several perceptual color spaces and warns about ineffective color combinations. (use) (source)
• d3-cam02: a D3.js module that implements the CIECAM02 and CIECAM02-UCS perceptual color spaces. (use) (source)
• d3-jnd: a D3.js module that lets designers check whether two colors are easy to tell apart, accounting for how discriminability shifts with mark size. (use) (source)

Beyond color, I studied how information retrieval techniques can aid the qualitative evaluation of visualization designs, how to design effective genome visualizations, and topics such as software visualization and crowdsourcing.