Stephen Grimes, Ph.D. Email & Phone Number

Boston, Massachusetts, Us

- Internal Python library development, package maintenance, CI testing and distribution- Data Engineering using Delta on Databricks with Starburst and Snowflake- Databricks administration on AWS with Terraform

San Francisco, California, Us

- ML Platform webservice using Ray + Kubernetes backend- AWS infrastructure using Terraform- Splunk Intelligence Assistant: SPL code generation using LLMs (Starcoder)

Seattle, Wa, Us

• As tech lead for MLOps “Model Infrastructure Layer” at Alexa Entertainment, designed and implemented a continuous integration/delivery (CI/CD) machine learning model pipeline to manage daily model re-training and batch inference of 10 machine learning (ML) models and dozens of ML workflows using AWS StepFunctions, SageMaker, Lamba, SQS/SNS, and EC2. • Led design reviews, provided design feedback to team members, and served as primary code reviewer for both engineering and science ML teams.• Authored an automated tool to ingest catalog entities from partner companies (Netflix, Spotify, Apple, Hulu, etc.) to test for broken natural language model interpretations for customer video, radio, music, and podcast requests. Corrected 40k defects monthly, generating $4k monthly in downstream customer revenue.• Architected and developed the Unified Fire TV-Alexa Dataset for analyzing Fire TV user click and voice interactions. Features extracted from the dataset enabled a clickstream model to propose speech recognition model corrections using subsequent Fire TV user text queries. • As a “Security Certifier” in Amazon Devices & Services, conducted architecture and threat model reviews for Alexa teams and performed manual and automated code reviews to ensure microservices securely handle Alexa customer data.• Acted as “away team member” in Alexa Speech to build a Lambda API service allowing Alexa Entertainment to test and safely release entity catalog model updates daily instead of monthly.• Languages: Java, Python, JavaScript/TypeScript• Workflow Orchestration: StepFunctions, Simple Workflow (SWF), Docker, Luigi• AWS: Lambda, SageMaker, Simple Workflow (SWF), EC2, ECR, SQS, SNS, S3, EMR, Redshift, CloudWatch, IAM, Polly, DynamoDB, Athena, Glue, SES• Additional Libraries: pyspark, pandas, numpy, jupyter, scikitlearn

Houston, Texas, Us

• Tech lead for NRG OneApp project, which migrated the legacy Reliant Energy mobile app backend to integrate with the NRG Energy consumer mobile app.• Increased use of serverless by migrating from Microsoft SQL Server and C# and to use python Lambda functions where possible.• Implemented app authentication using Cognito and Mobile Hub• Languages: C#, python, bash• DevOps: Terraform, CloudFormation• AWS Technologies: SQS, SNS, Lambda, Mobile Hub, IAM, Cognito

Radnor, Pa, Us

• As full stack engineer, built and maintained multiple microservices and APIs using Clojure and Javascript/Angular to support a business-to-customer text messaging app and admin web portal. • Contributed to greenfield platform rewrite from Clojure to Java while maintaining and building features for existing platform and migrating functionality to new platform.• Implemented rspec testing framework to harden platform and reduce burden on QA team.• Languages: Clojure, python, Java, ruby, JavaScript, Scala• AWS: EC2, RDS, S3, Redshift, Route53, SNS, SQS, IAM, VPC, Load Balancing, Auto Scaling• Tools and Libraries: AngularJS, iPython, jupyter, Ansible, Apache Kafka w/ Zookeeper, RabbitMQ, PostgreSQL, MySQL, Couchbase, Redis, NGINX, Twilio

Philadelphia, Pennsylvania, Us

• Supervised software development team supporting corpus and resource creation for DARPA-sponsored projects on machine translation, information retrieval, handwriting analysis, information extraction, image analysis, speech recognition, and language and speaker identification.• Managed team sprints using JIRA; introduced code reviews and experimented with pair programming and test-driven development to share knowledge and reduce bug fixes.• Wrote annotation GUIs using Python with Qt to record gold standard word alignments between parallel sentences for machine translation applications.• Bootstrapped manual word alignment annotation by pre-labeling using prior human annotations as training data, increasing annotation speed by 40% and saving ~$45k in project costs.• Designed and built widely-used English-Arabic Parallel Word Aligned Treebank and Arabizi-Arabic text messaging corpora for machine translation research.• Developed CodeIgniter/PHP/JavaScript web app for annotating handwriting collections.• Designed Mechanical Turk annotation workflow tasks to scale annotation delivery and reduce costs.• Designed database schema and XML/DTD transfer formats to capture human annotation of natural language for machine learning.

Philadelphia, Pennsylvania, Us

• Designed semantically intuitive database schema and XML/DTD/RNG formats to deliver human annotations of natural language for machine learning clients• Managed sanity checks and quality control to validate corpus releases prior to distribution to partners• Supported PHP/Codeigniter/Javascript web app for handwriting collection. Used segregated databases to protect Personally-identifiable Information (PII) from possible exposure. • Designed innovative Parallel Word Aligned Treebank corpus for machine translation research

Philadelphia, Pennsylvania, Us

• Wrote desktop GUIs using Python and PyQt Canvas to permit gold standard word alignment annotation between parallel sentences in two languages• Increased manual word alignment throughput by 10% using machine learning of previous annotations to suggest annotations to human annotators. Saved thousands of government research funds by eliminating need for redundant annotations• Presented research and domestic and international conferences and published results in books and conference proceedings

Burlington, Ma, Us

• Implemented quality control and enrichment of pronunciation lexicon using manual and rule-based strategies.• Scraped websites to add out-of-vocabulary words to pronunciation lexicon.

• Developed Dictionary Database and Automated Text Processor software programs to manage linguistic resources of Pawnee and Lakota languages• Wrote Adobe Flash language games for American Indian educational curricula. Game software served to elevate the status of the indigenous language as a modern language in the eyes of the community.• Created fonts and converted language data between font encodings• Designed ETL pipeline to move language lexicons between XML, FoxPro, Shoebox/Toolbox database formats

Bloomington, Indiana, Us

• Taught courses in Calculus and Precalc, Algebra, Discrete Mathematics, Statistics, and Language Analysis• Wrote curricula, designed quizzes and exams, held review sessions and office hours

• Ran expectation-maximization algorithm to build maximum-likelihood language models (LMs) for automated speech recognition. The LMs produced transcriptions of medical reports recorded into a Dictaphone.• Localized LMs to individual hospital sites by adding custom entities for each location.• Re-implemented language modeling tools written in Perl to C for performance optimization.• Curated medical data for training by writing tools to sanitize personally identifiable information (PII) of patients prior to tokenization and language modeling.

• Wrote code in GAP (Groups, Algorithms, Programming - a System for Computational Discrete Algebra) to perform research in computational group theory• Searched for high order groups to answer questions derived from the Burnside problem, which seeks to determine whether every finitely generated group in which every element has finite order must necessarily be a finite group. Burnside problem: https://en.wikipedia.org/wiki/Burnside%27s_problemGAP programming language: https://github.com/gap-system

Easton, Pa, Us

• Used masks to constrain the orbits of iterative functions to produce visualizations of chaotic attractors having each of the 17 possible planar symmetries. • Research resulted in several conference presentations and three publications. • Wrote code using J, a descendant of APL. J is useful for broadcasting scalar operations to arrays and matrices and ability to use diacritics to modify n-ary operators

Certificate, Business Essentials / Management And Finance

Bachelor Of Science, Mathematics And Computer Science Minor

Ph.D., Linguistics

Master Of Arts - Ma, Computational Linguistics

Master Of Arts - Ma, Mathematics