Zachary Pardos, Professor

Closed (1) Supporting undergraduate decision making with big data and adaptive technologies

Applications for fall 2021 are now closed for this project.

The AskOski project (askoski.berkeley.edu) is a research and development effort lead by Professor Pardos to transfer the latest in big data and machine learning research to the domain of course guidance in higher ed. The recommender system is live in-production at Berkeley and is being deployed at two other campuses this year.

There are two available roles, each with different tasks and learning outcomes:

(1) Machine learning engineer. This role involves optimizing and writing tests for existing machine learning models. Tasks include training predictive models and testing them on several datasets, increasing runtime efficiency, and working with researchers to integrate successful approaches in to the production system. Learning outcomes include familiarization with one or two modern neural network models (e.g., word2vec, RNNs, Transformer-based models) and learning how ML can be integrated into a live product.

(2) Front-end Developer. This role extends the functionality or robustness of the recommender system on the web-based front-end written in AngularJS and HTML. Tasks include adding additional features to the frontend, implementing UI improvements, and optimization of back-end calls and subsequent data processing. Learning outcomes include familiarization with modern web programming paradigms and exposure to a product development stack.

(3) Back-end Developer. This role extends the functionality or robustness of the recommender system on the Python-based back-end. Tasks include implementing new features for AskOski's search, plan, or explore functionalities and optimizing existing features. Familiarity with MySQL, Pandas, and Flask is encouraged. Learning outcomes include familiarization with back-end systems and programming, database management, and exposure to a product development stack.

(4) Pipeline Developer. This role involves designing and developing the machine learning and data preprocessing pipeline for AskOski. Tasks include generalizing data schemas for extending support for AskOski to a multicampus framework, streamlining and optimizing the machine learning model training by using prior trained models, and maintaining the pipeline for new frontend and backend feature additions. Learning outcomes include familiarization with Apache Airflow and real-world workflows for a production machine learning pipeline.

(5) Development Operations Engineer. This role involves researching approaches to improving the quality and scalability of AskOski’s system. Tasks may include (a) writing tests to ensure quality throughout the AskOski system (b) researching what AWS integration might look like (c) designing a multicampus deployment framework. Learning outcomes include familiarization with testing and CI/CD technologies including Docker, Kubernetes, and Github Actions.

Qualifications: Requirements: For position 1: Experience training neural networks in pyTorch or Keras For position 2: Angular JavaScript (for front-end) For position 3: Python and Flask, optionally Airflow and MySQL (back-end) For position 4: Data science / data engineering skills For position 5: Experience with AWS or testing or deployment frameworks

Weekly Hours: 6-8 hrs

Off-Campus Research Site: Remote

Related website: https://askoski.berkeley.edu
Related website: https://gse.berkeley.edu/zachary-pardos

Closed (2) Expanding a free content library for an open-source adaptive tutoring system

Applications for fall 2021 are now closed for this project.

Adaptive tutoring systems are designed to provide students in K-12 and intro college courses a personalized homework experience. This means giving the right problem to a student at the right time based given a continuous assessment of their mastery of a skill. At Berkeley, the ALEKS system is used to bridge gaps in Mathematics knowledge between high school and college. These systems are effective but costly and this cost represents an impediment to equitable education, since not all parents and districts can afford this type of product. Professor Pardos is working on an open-source version of this type of tutoring system. The code-base for the tutoring system has been established, but the content is currently limited. This project calls on students with interests in teaching, technology, and equity to draw on open license educational content on the internet (called open educational resources) to help build-out the content pool for this open-source system.

The role would involve (1) transcribing educational content from the web into a Google spreadsheet format amenable to the adaptive tutoring platform (2) periodically creating new tutorial content and (3) editing testing, and quality checking educational content transcribed by a network of qualified volunteers.

Learning outcomes include (a) understanding what adaptive tutoring is and the various components of this educational technology and (b) learning effective pedagogical/teaching practices in an increasingly digital world.

Qualifications: Applicants must have an interest in education, technology, and equity. Teaching or tutoring experience in STEM is recommended.

Weekly Hours: 6-8 hrs

Off-Campus Research Site: Remote

Related website: https://math.berkeley.edu/courses/choosing/aleks
Related website: https://gse.berkeley.edu/zachary-pardos