top of page

Jake Metzger

AI Researcher. ML Engineer. Philosopher.

  • LinkedIn
  • Medium
  • orcid.logo.icon.svg

Welcome to my website!

​

I am an experienced AI researcher and machine learning engineer interested in bridging the gap between the scalable application of next-gen AI technologies and the moral and intellectual dimensions of human decision making. I have keen research interests in ML-driven probabilistic epistemology and applied ethical / responsible AI.

​

Here, you can find my resume/CV outlining my work and interests and my blog where I'll post some of my thoughts.

​

Note: None of the thoughts or expressions on this site are reflective of any opinions or positions of any of my employers or affiliations at any time.

​

Resume

RESEARCH INTERESTS

Data Science and Machine Learning

Deep Learning. Simulation and Decision Science. Bayesian Statistics. Uncertainty Quantification. Causal Inference.

Philosophy of Science

Computation and Intelligence. Logical and Statistical Inference.

Ethics and Metaethics

Ethical and Responsible AI. Epistemic Norms, Virtues, and Responsibilities. Expressivism and Cognitivism.

Epistemology

Bayesian and non-Bayesian probabilistic reasoning. Decision theory.

PROFESSIONAL EXPERIENCE

Associate R&D Scientist - Machine Learning

Accenture Labs - Systems and Platforms

San Francisco, CA

  • Research and proof-of-concept development in hierarchical Bayesian modeling, uncertainty quantification, and targeted learning as applied in several concrete domains, including robust inventory management for supply chains and efficient, multi-constraint demographic targeting in medical trials.

  • Enablement of epistemically-aware, graph-based, robust ampliative inference for, for example, intelligent digital twin systems.

  • Active member of TEthER: Tech-Ethics Engaged Researchers. Lead several discussions regarding ethical applications of contemporary AI/ML, especially new GenAI.

  • Papers:
    • 2023 IEEE ICSC Conference Paper - Ontology Modeling for Probabilistic Knowledge Graphs
    • 2023 IEEE ICRA Conference Paper - Online Tool Selection with Learned Grasp Prediction Models (see OSARO)
    • 2023 IJSC Journal Paper - A Bayesian Approach to Constructing Probabilistic Models from Knowledge Graphs
       
  • Patents Pending:
    • 2023 - System for Probabilistic Modeling and Multi-Layer Modeling for Digital Twins
    • 2023 - Survival Prediction and Censored Demand Sensing for Inventory

2021 - Present

Machine Learning Engineer

OSARO, Inc.

San Francisco, CA

  • Co-development and implementation of ML technical and data requirements across engineering teams.

  • Improved quality of data ingested for ML training/prediction by implementing data quality safeguards.

  • Integration with high-throughput gRPC prediction and in-house analytics services.

  • Research and development of production-ready ML/DNN models for robotics applications at scale including supervised, unsupervised, self-supervised, generative, and reinforcement methods using multidimensional and multimodal vision, sensor, and motion data.

  • Improved on-site pick-and-place performance by 14% with occlusion-aware, hybrid pick prediction model.

  • Lead ML developer of box-pick solution, combining OSARO's registrationless pick technology with domain-specific awareness, improving pick quality and single pick success.

  • Statistical and off-policy analysis of model performance for cross-domain generalization.

  • Patents Pending:

    • 2022 - Automated Robotic Tool Selection

2019 - 2021

Data Scientist

Accenture - Liquid Studio

San Francisco, CA

  • Built rapid prototypes and proofs of concept to validate cutting-edge ideas and technologies for market.

  • Significant experience with natural language processing and conversational AI design, which has been taken from prototype into production at scale.

  • Focuses on explainable AI (XAI) and providing scalable, "glass-box" reasoning and dynamic user interaction capabilities to AI systems.

  • Deployment to heterogeneous cloud platform (AWS, Google Cloud, Microsoft Azure)​

  • Oct 2018 - Hands-on Conversational AI Lab faculty for Technology Architecture Workshop 2018

  • Dec 2018 - Promoted from Adv. App Engineering Sr. Analyst to Adv. App Engineering Specialist

  • Dec 2018 - Team named FY19 Q1 Demo Award winner for development of Flare360, a next-gen, semi-stochastic conversational agent demonstrated with industry use cases in finance, healthcare, and utilities.

  • Feb-Jun 2019 - Team named Top 5, then Top 3, global Breakthrough panelist in Greater Than Awards for development of onyxAI, a novel explanation-driven cognitive platform (Patent Issued July 2023)

  • Jun 2019 - Transitioned from Accenture Liquid Studio to Accenture Labs as Technology Research and Development Specialist. Researched novel solutions to open-world learning.

​

  • Patents Granted:

    • 2019 - Explanation-driven Interactive Recommendation Engine (Issued July 2023)

2018 - 2019

Artificial Intelligence Analyst

Accenture - Liquid Studio

Phoenix, AZ

  • Rapid design, prototyping, and deployment of intelligent, next-gen natural language agents using Amazon AWS, Google Cloud, and on-premises platforms in hybrid enterprise architectures.

  • June 2018 - AWS faculty for annual TechStars conference. Led hands-on rapid learning labs for Amazon Lex and Amazon SageMaker.

2017 - 2018

APPLIED SKILLS & EXPERTISE

AI/ML

Supervised, Unsupervised, Clustering, Deep Learning, Reinforcement Learning, Transfer Learning, Bayesian Modeling, Model Ensembles: Bagging/Boosting, Model Evaluation and Optimization, Off-policy Evaluation, Knowledge Graphs / Ontologies

Statistics and Data Science

Descriptive, A/B and Hypothesis Testing, Bayesian Inference, Predictive Modeling, Uncertainty Quantification

Programming

Python, Java, Scala, Javascript, Nim, gRPC, SQL, SPARQL, Prolog / miniKanren

Libraries and Frameworks

Tensorflow, Keras, scikit-learn, xgboost, Apache Spark, hyperopt, PyMC, PyTorch, Pyro, Tensorflow Probability, PyTensor, Aesara

Platforms

Amazon Web Services: EC2, S3, Sagemaker, Lex, Lambda, Connect, DynamoDB, Neptune, Cloudformation

Google Cloud: Compute, Dialogflow, Firebase, Cloudfunction

Containers: Docker, Podman, Vagrant

SELECTED APPLIED AI PROJECTS AND RESEARCH

Censored Cumulative Demand Estimation for Data-Driven Inventory Management

Accenture Labs - Systems and Platforms

San Francisco, CA

  • Designed and implemented a dynamic machine learning system that utilizes censoring-aware analysis from sales transactions to provide unbiased, on-demand inventory forecasts to reduce inventory costs and maximize product availability.

  • Extended work to include multiple, implicit censoring sources in sales transaction data, including demographic-specific substitution and complementation.

  • Further extended work from optimal inventory management to optimal pricing via unconstrained, demographic-specific cumulative demand forecasting.

  • Patent pending

Orion - Probabilistic Knowledge Graph Platform

Accenture Labs - Systems and Platforms

San Francisco, CA

  • Designed and implemented a probabilistic reasoning system for knowledge graphs, extending inference capabilities from logical/semantic reasoning to structural-probabilistic.

  • Further extended reasoning capabilities from factual to practical inference, optimizing decision making under uncertainty using Bayesian decision theory.

  • Applied to medical drug trial use case for optimal site selection, improving trial completion rates and statistical data quality.

  • Conference and journal papers published

  • Patent pending

  • Published two introductory blog posts on Labs Notebook (Medium)

Adaptive Grasping and Tool Changing for Robotic Picking

OSARO, Inc.

San Francisco, CA

  • Proposed and implemented context-aware, adaptive methods for dynamically improving pick performance during a robotic pick mission by using MPC-based future planning.

  • Improved success over existing solution by 19%.

  • Conference paper published.

  • Patent pending.

Box Object Prior for Registrationless Robotic Picking

OSARO, Inc.

San Francisco, CA

  • Developed box-specific object prior model to complement OSARO's object-agnostic registrationless pick modeling for packing use cases. Utilized RGBD-based instance detection to prevent product pick and place failures not captured by extant reward signals.

  • Improved success over existing solution by 14%.

XAI Recommendation System (onyxAI)

Accenture - Liquid Studio

San Francisco, CA

  • Built explanation-driven, human-in-the-loop recommendation system using state-of-the-art explainable AI (XAI) research from Accenture Labs (Ireland) applied to hospitality industry use case.

  • Utilized sentiment analysis, knowledge graphs, and dynamic and corrigible human-in-the-loop recommendations.

  • Served as project's AI SME for semi-finalist and top-3 finalist rounds of Accenture Greater-Than Awards (out of 180+ global contenders).

  • FY19 Greater-Than Awards Top 3 Global Finalist, Breakthrough category

  • Patent granted for underlying technology (see above)

Flare360 Intelligent Customer Agent

Accenture - Liquid Studio

San Francisco, CA

  • Core designer and developer of Flare360 - a next-gen, semi-stochastic conversational agent with use cases in finance, healthcare, and utilities.

  • FY19 Demo Award Winner

Data and Algorithm De-Biasing

Accenture - Liquid Studio

San Francisco, CA

  • Developed applied proof-of-concept demonstrations in collaboration with Dr. James Zou's Stanford University AI research group targeting data and algorithmic bias, black-box model auditing, data quality, and contrastive analysis.

  • Provided early-stage demo for integration with Accenture Data Intelligence Suite (DIS)

EDUCATION

University of Illinois - Springfield

Master of Science in Computer Science

Emphasis in Data Science and Machine Learning

Springfield, IL

2017

University of Arizona

Bachelor's Degree in Mathematics, Philosophy (Dual)

Tucson, AZ

2010

CONTACT

Email

j.metzger@arizona.edu

  • LinkedIn
  • Medium
  • orcid.logo.icon.svg

Do not contact me to sell me things.

bottom of page