The Manifold Labs team partners with clients to build large-scale data and machine learning systems and to invent novel applications of machine learning. We bring the curiosity and mathematical depth of a research lab and the pragmatism and engineering depth of a technology company. We create impactful technology on a foundation of true partnership rather than vendorship. We constantly explore leading-edge methods and tools so we can be the best partner for future clients and we share our insights in top venues.
Built fit-for-purpose machine learning infrastructure using public cloud and open source components to streamline experimentation, deployment, and monitoring across varied use cases and teams. Built the first customer-facing applications on the platform, using billions of data points to create 300 patient features. Enabled data science teams to work faster by taking care of a lot of heavy lifting and catalyzed a new “data product” mindset.
Conducted multiple user interviews across Commercial, Medical, and Clinical leaders – discovering the key business priorities for each organization and creating product roadmap. Built an intuitive and engaging user interface. Built modern data infrastructure for data aggregation, including automated data quality checks and new sources of data to be accessible to sales and marketing in a compliant and automated way.
Conducted user interviews across sales organization to understand sales motion, decision rules, data and information flows, and seller onboarding. Designed a “single pane of glass” data and ML app. Built a customer data platform to aggregate relevant data from multiple systems. Trained and deployed a machine learning model to rank accounts by predicted likelihood to purchase in a given quarter. Increased efficiency of sellers, of seller onboarding, and of cross-sell and up-sell attributed with >10% increase in revenue.
Conducted user interviews across organization to discover business priorities and workflows. Built an intuitive and engaging user interface with multiple data layers. Built modern data infrastructure.
Experimented with deep learning architectures for accelerating inverse modeling in pharmacokinetics / pharmacodynamics (PK/PD). Advanced ability to do thousands of experiments rapidly and train architectures with low error over a large range of parameters.
Rapidly developed multiple concept prototypes. Downselected to a first use case to bring to market, with a longer term platform vision. Developed product experience with early customers.
Explored 7 years and 10 terabytes of data on acute hospital discharge and referral data. Set-up HIPAA-compliant cloud environment, designed for rapid AI/ML experimentation. Prepared the data for modeling. Developed an ML model to predict whether a patient was likely to need skilled nursing care or home care after hospital discharge and the associated readmission risk. Demonstrated how model explainability helps end-users accept an ML model and see how the insights were generated.
Use computer vision after construction is complete for major retailers so that they can track resources in stores (e.g. shelves in drug stores, chairs and tables in chain restaurant)Reimagined store resource management by utilizing computer vision to create more efficient project processes and asset management for major retailers (e.g. shelves in drug stores, chairs and tables in chain restaurant).
This collection of essays, papers, and conference talks reflects our pragmatic approach to machine learning.
