🇻🇮🇩🇴 Jazmia Henry (She/Her)

NeurIPS

Paper "Scaling Laws or The Law of Diminishing Returns: How Scaling Low Quality Data Degrades Language Model Performance" accepted for poster presentation with a top 5% rating and short listed for oral presentation consideration at Black in AI workshop. Presentation at NeurIPS 2023.

Accepted Paper (Top 5% Rating) 

Founded in 1987, the Conference on Neural Information Processing Systems (abbreviated as NeurIPS and formerly NIPS) is one of the most prestigious and competitive international conferences in machine learning and computational neuroscience.

Reviewer - Datasets and Benchmarks Track

Keynote Presentation: https://jazmiahenry.com/my-presentations/neurips-keynote-jazmia

Keynote Speaker - Global South in AI Affinity Workshop

Microsoft

Microsoft MLADS is the internal Machine Learning, AI and Data Science conference where employees present tutorials, presentations and papers. I serve as a Reviewer.

Machine Learning, Data Science and AI Conference (MLADS) Reviewer 

Senior Data and Applied Scientist for ISE, FSI team. Focused on bringing LLM solutions using Azure's many Generative AI offerings including Azure OpenAI, Microsoft Fabric and Azure Machine Learning Studio. 

Formerly a Senior AI Applied Engineer building Autonomous AI Solutions for Industrial Partners using the Reinforcement Learning platform, Microsoft Bonsai, from August 2022 - July 2023.

Senior Data and Applied Scientist, Data and AI 

Architected scalable Machine Learning and Reinforcement Learning models, driving productivity and cost savings for industrial clients.
Product owner for Azure Plato, an open-source toolkit, facilitating strong feedback loops with partners and customers.

Senior Applied AI Engineer, Project Bonsai

McGraw Hill

Writing a chapter in the upcoming textbook, “Mitigating Bias in Machine Learning”, to be released Summer 2023.

Author

EYPHKA

Mentor founders from historically excluded backgrounds to navigate the landscape of startup formation, funding raises, hiring, Data strategy and Machine Learning infrastructure.

Startup Mentor 

www.eyphkaai.com

Thesis: 
What would the future of computing look like for AI? Since Amazon Web Services created Elastic Cloud Compute in 2006 for enterprises to be able to own and build applications without the need of physical servers housed on premises, Machine Learning became more mainstream as companies could now build ML applications with less overhead. Now, all major Tech companies use cloud based servers for application building and advanced companies use EC2s and S3 for Cloud Big Data storage and model compute.

	As IoT devices grow in popularity, edge computing is changing the way we approach compute with edge Machine Learning engineers training smaller models on edge layers and deploying model outputs using dynamic cloud orchestration that separates gateway functionality (normally in some breakout point) from model output consumption (normally on the network’s edge). This adaptability to a smaller device that owns more data and needs quicker response times has evolved what we consider to be optimal compute. 

	Decentralization through Blockchain has the power to take this idea even further and led us into a new realm of what’s possible with Machine Learning.

CEO and Founder

Correlation One

Lead Instructor - Data Science and Machine Learning 

Stanford University

Building an open source African American Vernacular English  corpora package to mitigate bias in downstream NLP tasks using Deep Learning Techniques such as few shot and zero shot learning and transfer learning.. 
Participates as a panelist, technical contributor, and speaker at academic and public companies.

Major Accomplishments and Publications:
“Mitigating Bias During the MLOps Lifecycle”. MIT Press. 2022. Pending Publication. 
“Practical Challenges to Achieving Algorithmic Fairness in Industry (And What to Do About Them)”  August 2021. Amstat. Panelists and Other Contributors: Sarah Cox, Coinbase; Samsara Counts, Amazon; Dr. Kalinda Ukanwa, University of Southern California
“AAVE: A Collection of AAVE by Black Speakers for NLP Data Supplementation”. 2022. NeurIPS. Pending Submission.

Technology Fellow: NLP

The Motley Fool

Founding member of the Machine Learning Team at the Fool. Lead team hiring, Company-wide AI strategy, and Machine Learning Model Building. Lead Machine Learning Modeler that built and deployed large scale Recommendation Engines, Deep Learning Models, Regression Models, and Quantitative Models.
Assisted in Cloud Migration from On-Prem servers and integrated Streaming servers into a Data Lakehouse structure for enhanced ML performance, Data democratization, and Data Enablement.
Directly managed a team of ML Engineers, Data Scientists, and Project Managers. Collaborated with Business Intelligence and Data Engineering Department Heads to form a collective Data team of over 30 Data Engineers, Business Intelligence Analysts, and Machine Learning Practitioners to build, deploy, and manage robust ML models for millions of Motley Fool users.

Major Accomplishments:
Built and Deployed a Hybrid Recommendation System using the Alternating Least Squares Matrix Factorization technique using Spark.
Worked with leadership to establish AI Ethical Standards and brought on consultants to create Ethical Product Review Standards for AI models and with business leaders to create company wide AI and Data literacy programs.

Head of Machine Learning

SIID Technologies

Co-Founder and Advisor 

Morgan Stanley

• Relationship Manager that communicated model performance, success metrics, predicted ROI, and model scope to senior executives, junior developers, and non-technical management located domestically and abroad.
• Perfected ETL process for large and distributed data sets through Hadoop and Spark using SQL and
Pyspark. Built queries, reports, and models using R and Python. Used Tableau to visualize findings through dashboards to make business recommendations to senior shareholders.

Data Strategist

FightPandemics

FightPandemics is an altruistic platform created in the wake of COVID-19 that connects people and organizations in need of help with those that can provide it. In fostering these vital connections, we empower communities to respond more quickly and effectively to the crisis. Our vision is to create a world where humans are more resilient to emergencies.

We are a global group of volunteers that came together to make a difference and are actively looking for other highly-dedicated individuals to join us.

Data Manager Volunteer

As a Senior Data and Applied Scientist at Microsoft, I work with cutting-edge technologies and solutions in the fields of Generative AI and Natural Language Processing. I have 10 years of experience in building Machine Learning in different industries: training Autonomous AI Solutions for Industrial Partners using the Reinforcement Learning platform, Microsoft Bonsai, bringing Large Language Model solutions using Azure's many Generative AI offerings, including Azure OpenAI, Microsoft Fabric, and Azure Machine Learning Studio, building Hybrid LLM/Recommendation Systems to provide content recommendations to users at the Motley Fool, making financial models for UNHW clients at Morgan Stanley and more.

I've led and mentored Data Scientists, Machine Learning Engineers and Product Managers passionate about building AI that makes people's lives better. I've spoken in front of audiences of global leaders at International conferences and Seminars about the future of Machine Learning and LLMs. I've built products that work and products that don't work that use AI. I've written papers and a textbook and now I share my knowledge with the world. 

I'm currently a PhD Candidate at the University of Oxford, where I study Social Data Science and contribute to the research and development of ethical and responsible AI. I am passionate about advancing the frontiers of AI and Data Science, sharing my knowledge and insights with diverse audiences, and collaborating with peers and experts across disciplines and domains.

I've done a lot of cool stuff and would love it if you would follow me on my journey. If you would like to learn more about my work, explore my website: jazmiahenry.com/
And Oxford profile: https://www.oii.ox.ac.uk/people/profiles/jazmia-henry/

Senior Applied AI Engineer, Autonomous Systems

The AI Strategy I enacted at The Motley Fool in 2021 was cutting edge…here’s how we used AI back then to solve actual user problems. 

Our amazing Head of User Research Helen E. Fake, Ph.D. identified a core problem on our Motley Fool website that was affecting customer satisfaction and engagement. People were having a hard time navigating the Fool’s large IP and, as a result, felt frustrated and would frequently bounce from our Financial advice offerings. 

So the ML team with full faith of our CTO at the time, Nathan McMahon, built something that hadn't been built before.

Using the tripartite method described in the book Value Systems Design, l designed a hybrid Recommender System that ingested word embeddings of all of the content from Motley Fool in a set timeline, used cosine similarity on those embeddings and features of users grouping them together through a clustering method based on their engagement habits and reading patterns, and built an Alternating Least Squares matrix factorization model to recommend the articles people would like the most while not falling into the trap of perpetuating bias.

I built a RecSys model and Beegee Alop led the deployment and monitoring of the model. Our Product and Content leads were the Subject Matter Experts lead by Katie C.. Our Data Engineers led by Motley's current Chief Data Officer, Nicole Naworal, ran and stored the model in batch. Our former CTO, Adam Hill, created a middleware that secured the entire system and communicated between front and backend commands.

We ran Blue/Green experimentation on the model's effectiveness vs what we called the "status quo"- keeping the website the same for some users- and a "basic tech" model - giving people the newest articles to read next with no ML powered recommendation. We used Canary deployment to slowly unveil the feature. Our Data Analysts led by the Head of BI, Lauren Danker Lutes, measured the success of our experiment against our KPIs.

Bruce Hoffman led our Responsible AI efforts ensuring our model was ethical and our use of it was transparent to users. Our Legal team made sure we properly handled data and considered our regulatory compliance limitations.

The ML model increased engagement across the board but particularly amongst women and people of color on the site. It reduced page bouncing significantly over the status quo and basic tech offering. 

Nowadays, this would be called Generative AI with a Recommendation System built atop to enhance performance.

This is what user centric, value sensitive Generative AI design looks like in action. It takes everyone- every team working together doing what they do best to make it happen. 

I knew that back in 2021.

If you want to build something that has never been done, you have to create novel standards with diverse people, minds and backgrounds to make it happen.