Suggestions
Daniel Smilkov
Software Engineer at Google
Daniel Smilkov is a machine learning and visualization expert with a distinguished career in technology and research. He was a Staff Software Engineer at Google from April 2019 to April 2023, working on projects related to Responsible AI and TensorFlow.js.1 Prior to that, he worked as a Software Engineer at Google from September 2014 to April 2019, focusing on the intersection of machine learning and visualization as part of the Google Brain team.1
Smilkov co-created TensorFlow.js and Know Your Data, and was involved in PAIR (People + AI Research) initiatives.1 He holds a Master of Science degree from the MIT Media Lab, which he completed from 2012 to 2014.1
Currently, Smilkov is a co-founder of Lilac AI, a company he started in April 2023 that focuses on analyzing, structuring, and cleaning unstructured data with AI.1 Lilac AI was recently acquired by Databricks.3
Throughout his career, Smilkov has made significant contributions to the field of machine learning and data visualization, with numerous publications and patents to his name.2 His work includes research on smoothing gradient techniques, visualization of deep learning models, and the development of human-centered tools for medical decision-making.2
Highlights
Excited to announce Lilac, a new Open Source product to analyze and refine text datasets.
At Lilac AI, our mission is to make unstructured data visible, quantifiable, and malleable. This will lead to:
- Higher quality AI models
- Better actionability when AI models fail
- Better control and visibility of model bias
Live demo of the OpenOrca dataset: https://lnkd.in/dydRjadB Documentation: https://lilacml.com
Lilac lets you:
- Visually explore datasets
- Enrich with structured metadata using "Signals", e.g. near-dup and personal information detection. This allows us to compute statistics, find problematic slices, and eventually measure changes over time.
- Create and refine "Concepts" which are embedding-based AI models that can be built via the UI and can be used to find and score text from datasets that match a semantic concept you have in your mind.
- Download the results of the enrichments for downstream applications.
🔒 We are focused on making the product fast and useable on-device. We are using on-device Open Source embeddings, like GTE. If you prefer to use a more powerful embedding, like OpenAI, PaLM, or Cohere, you can use your own API key.
We believe the best way to improve the quality and malleability of datasets is a hybrid of a Python API and a visual UI, built Open Source.
We also believe a community is the best way to grow useful signals and concepts. Find us on GitHub if you have issues or suggestions: https://lnkd.in/dWSH-C-P
