Boris聽Dayma
Founder at Meridian Physician Search 馃┖
United States
Summary
馃ぉ
Rockstar馃帗
Top SchoolBoris Dayma is a founder and machine learning engineer with 11 years of experience building AI tools and startups, currently leading Craiyon and Meridian Physician Search. He combines hands-on ML engineering鈥攃ontributions to high-profile open-source projects like DALL路E Mini, Hugging Face Transformers, fastai, and Weights & Biases integrations鈥攚ith product-minded leadership in niche markets such as outpatient psychiatry recruitment. His work spans model training pipelines, data engineering for massive image datasets, and practical experiment tracking, reflecting a focus on reproducible ML workflows and developer UX. Trained as an engineer in Europe and Brazil, he blends rigorous academic foundations with production-grade software craftsmanship and a history of improving observability and experiment logging across major ML frameworks. Notably, he implemented core training and logging features in widely used repos, helping democratize image generation and model experimentation.
12 years of coding experience
10 years of employment as a software developer
Centralien graduate engineering degree (Master of Science), Centralien graduate engineering degree (Master of Science) at Centrale M茅diterran茅eLyc茅e Pierre CorneillePontifical Catholic University of Rio de Janeiro
French, Chinese, Spanish, Portuguese, English
Github Skills (69)
checkpoint10
python10
image-processing10
ui-components10
testing10
flax10
datasets10
mlops10
deep-learning10
trainings10
sequence-to-sequence10
streaming10
javascript10
openai-api10
optim10
Programming languages (10)
Github contributions (5)
borisdayma/dalle-mini
Jul 2021 - Jan 2023
DALL路E Mini - Generate images from a text prompt
Role in this project:
Back-end Developer
Back-end DeveloperContributions:4 releases, 14 reviews, 809 commits in 1 year 6 months
Contributions summary:Boris's primary contribution was the implementation of a new script, `run_seq2seq_flax.py`, for the DALL路E mini project. They integrated functionalities to process image encodings and captions, indicating involvement in the core sequence-to-sequence model pipeline. Subsequent commits involved adjusting the script for different data sources and functionalities like incorporating AdamW and Adafactor optimizers, suggesting the development and refinement of the training process. Further commits showed improvements to the training, like logging interval, and handling the number of steps for learning rate.
pythondall-etext-to-imagepromptdalle
rom1504/img2dataset
Aug 2021 - Jul 2022
Easily turn large sets of image urls to an image dataset. Can download, resize and package 100M urls in 20h on one machine.
Role in this project:ML Engineer & Data Engineer
ML Engineer & Data EngineerContributions:12 reviews, 21 commits, 10 PRs in 10 months
Contributions summary:Boris primarily focused on enhancing the `img2dataset` project to handle various data formats and improve image processing capabilities. Their contributions include adding support for TSV, TFRecord and different image encoding formats such as PNG and WebP, thus extending the project's flexibility for diverse datasets. They also implemented image resizing features and filters, including handling transparency and filtering images based on size and aspect ratio, enhancing the image dataset creation pipeline. They also incorporated W&B logging and introduced improvements to the code, such as handling transparency and accepting numpy arrays, as well as custom timeout.
image-urldeep-learningdatasetbig-dataimage
Find and Hire Top DevelopersWe鈥檝e 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.