Sam Havens

Writer is an AI writing assistant used by brands like Twitter, Intuit, and Accenture. The NLP team uses a microservice architecture based on Kubernetes, FastAPI, HuggingFace Transformers, NVIDIA Triton, and ONNX.

• Responsible for NLP from research to operations, including >25 microservices
• Trained an encoder/decoder Grammar Error Correction model using novel synthetic data techniques, which outperformed an open-source baseline by 130%
• Used NVIDIA Triton and ONNX to serve a character-based transformer spelling correction model while keeping inference latencies below 300ms at 50 req/s

• Reduced mean service latency from >1.5s to <300ms.
• Grew team from 2 to 6, while improving onboarding effectiveness (time to first commit: from weeks to < 1 day).
• Implemented classification and seq2seq models in spaCy, Flair, and Marian with aggressive latency requirements.
• Responsible for all ML Ops. Made models available using modern async/await Python, and Docker/Kubernetes/PubSub, with some help from Bash and Jenkins.

• Created a suite of tools for automotive OEMs and dealers to manage chatbots for their brand on web, chat, and voice platforms.
• Used Docker/Kubernetes to operate services written in Node.js, Elixir, and Python, and models made in FastText and TensorFlow.
• Engineering team grew from 4 to 16 during my tenure. Established a strong engineering culture of testing, code reviews, pair programming, and mentorship.

• Created a car comparison shopping tool using React, ES6, Webpack, and MaterialUI.
• Built a conversational agent with a Node.js/Docker backend and NLP services in Python using FastText, NLTK, and Gensim.

• Full-time math and science teacher from 2007-2014.
• Current board member.

• Implemented a probabilistic lead-scoring model and built an MVC lead generation web app with a LAMP backend and d3.js + jQuery frontend

• Bianchi Outstanding Graduate Student research award, and IRIS Fellowship recipient.

• LEAPS Fellow