top of page

What's a S.M.E.?

In machine learning, a "SME" or "subject matter expert" play a crucial role in various stages of the process. They are typically individuals who possess domain-specific knowledge and insights (i.e. underwriter in insurance, doctor at a hospital, master mechanic at an auto repair center) and collaborate with DSs (data scientists) and MLEs (machine learning engineers) to develop accurate and effective machine learning models. these experts, or SMEs, will have varying degree of tech understanding and knowledge from a non-user (perhaps uses a "dumb" phone) to someone who is up-to-date with all that is happening in the tech space.

We needed to create an interface that was easy to use and understand that had them focus on the data and not the bits and pieces of how to get from annotation to model creation for training purposes and iteration.

They're busy people who are sharing their knowledge wealth so that the platform is able to give better models and metrics for future machine learning projects.

Overall, SMEs play a critical role in bridging the gap between machine learning techniques and the real-world domain they are applied to. their expertise helps ensure that the models are accurate, relevant, and aligned with the specific needs and challenges of the given domain.

We did that by harnessing the power of the LLMs (large language models) and foundation models to empower SMEs to quicker, more accurate, efficient labeling of datapoints to kickstart programmatic labeling (annotations) through the iterative loop.

I also made this a little fun by including some LotR (Lord of the Rings) references for this presentation purposes only but the structure and how we approached assisted annotations is real and in the current platform (or resourced for development in the near future).

LLM

Leveraging LLMs, the user is able to quickly go through their assignments by prompting the platform to search, annotate, test, and validate their work while work for the admins and reviewers are reduced by confirming across multiple annotators.

MMM and Computer Vision

This interface is also able to handle MMM (multi-modal models). In other words, be able to take images, video, audio as inputs, and be able to detect items, elements within the frame based on foundation models.

Cross User Collaboration and Foundation Models

The annotation workspace is only one part of the platform. Data scientists, admins, and reviewers all use the platform, working together to build efficient, accurate, and fast models by iterating, testing, and validating.

S.M.E.agol

S.M.E.agol

Supercharging our precious with LLMs

bottom of page