Spencer Poff
Member Of Technical Staff at Harvey
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolSpencer Poff is an NLP engineer and product-minded technologist with 11 years of experience building production chatbots, ML infrastructure, and consumer-facing apps from startups to Meta/FAIR. Currently a Member of Technical Staff at Harvey, he previously tech-led chatbot teams at Meta contributing to the BlenderBot series and publications at ICLR, ICML, and EMNLP. He blends deep research sensibilities (Google Scholar presence) with hands-on engineering—debugging token-level loss in ParlAI and adding temperature decoding to improve generative outputs. A serial founder and builder, Spencer has launched AR field-service tooling and language-learning products, giving him a rare mix of product intuition and systems-level execution. Based in San Francisco, he pairs academic rigor with pragmatic product delivery, often surfacing subtle model-evaluation improvements that accelerate debugging and model iteration.
11 years of coding experience
11 years of employment as a software developer
Mechanical Engineering, Mechanical Engineering at UC Berkeley
Stackoverflow
1,640reputation
267kreached
6answers
4questions
statusbar
top-5%
height
top-5%
Github Skills (20)
pytorch10
python10
transformer-models10
ios9
statusbar9
height9
machine-learning9
nlp9
generative-model9
debug8
debugging8
shell6
frame6
adb6
uilabel6
Programming languages (4)
Github contributions (5)
facebookresearch/ParlAI
Nov 2019 - Aug 2022
A framework for training and evaluating AI models on a variety of openly available dialogue datasets.
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:88 reviews, 56 commits, 83 PRs in 2 years 9 months
Contributions summary:Spencer focused on debugging and improving the token-level loss calculations within the `parlai` framework. They added the ability to output and visualize token losses for generative models, aiding in debugging. Additionally, the user implemented and tested a new temperature decoding option within the `torch_generator_agent`, enhancing the model's output generation capabilities. Code changes are focused on model behavior, evaluation, and extending the core ParlAI functionality.
nlpdeep-learningdatasetmachine-learningtraining
spencerp/dotfiles
May 2018 - May 2024
Contributions:6 pushes in 6 years 1 month
xtermosxvimzsh-pluginbash
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.