Jesse Hamer Email & Phone Number

Iowa City, Iowa, United States

I lead research into new features and algorithms for Finetune's two AI-powered products, Generate (an AI-powered educational content generation system) and Catalog (an AI-powered educational content tagging system).

Design, experiment with, and prototype AI/ML models which drive functionality of innovative ed-tech products.Coordinate with AIRD team to set research objectives, and serve as interface with Dev team to help bring research into production.Major Projects:- Generate: AI-driven content generation for assessment and learning.- Catalog: AI-driven taxonomic tagging of assessment and learning content.

Design, experiment with, and prototype AI/ML models which drive functionality of innovative ed-tech products.Major Projects:- Generate: AI-driven content generation for assessment and learning.- Catalog: AI-driven taxonomic tagging of assessment and learning content.

Iowa City, Iowa, United States

Worked on AI-Machine Learning Team to rapidly ideate, prototype, experiment, and present AI and ML algorithms in order to solve problems and innovate in educational technology and measurement. Experimented heavily with state of the art transformer-based NLP algorithms.- Used Universal Sentence Encoding and TensorFlow to train deep neural network for speech behavior tagging. Improved prediction accuracy by +3.7%.- Wrote custom ALBERT models using the Transformers package and PyTorch. Improved accuracy on SQuAD2.0 dataset by +0.4% above baseline ALBERT model.

Kansas City, Mo

Consulted with legal timekeeping company Ping; developed a topic modeling pipeline from public record court documents to facilitate a fixed-rate billing model for legal work.1) Analyzed 3GB of raw text data from The Caselaw Access Project (parsed from compressed JSON files into a PostgreSQL database) using Python NLP packages Spacy for preprocessing and Gensim for LDA topic modeling.2) Validated topic model using perplexity, coherence, and a custom metric which associates topic similarity to degrees of separation in the citation network of court documents.3) Delivered final topic modeling pipeline with test coherence of 0.53, from baseline of 0.36.4) Delivered 31 Jupyter notebooks detailing experimentation with 5 preprocessing routines, 4 input datasets (controlling for different things like jurisdiction or decision date), 2 dataset sizes, and exploration of "legal stopwords".

Iowa City, Iowa Area

This competitive fellowship allowed me to focus on writing and defending my thesis in the final semester of my doctoral program.

Batavia, Il

Analyzed structural features of 80k+ CNNs (selected using an evolutionary algorithm) trained on 6.66GB of image data to predict which models would perform best on image recognition tasks on photos of particle interactions captured at Fermilab.1) Engineered 30 numeric features from semi-structured Caffe CNNs (stored as protobuf files) using Python. 2) Extracted features from 25GB of data by submitting batch jobs to a custom Singularity container on Fermilab’s Wilson HPC cluster.3) Discovered (using linear regression) that 13 of the engineered features and 41 of their interactions were significant in terms of their influence on task performance.4) Visualized CNNs in low-dimensional space with PCA; discovered that best performers cluster densely, while poorer performers disperse along three distinct trajectories.

1) Have taught a wide variety of coursework, ranging in difficulty from low level undergraduate (college algebra), to calculus, to graduate level (topology).2) Earned Graduate Certificate in College Teaching from College of Education.

Doctor Of Philosophy - Phd, Mathematics, 3.98

Studied low-dimensional topology, knot theory, braid theory, contact geometry. Used Python and SageMath to explore knot data and validate.

Master Of Science - Ms, Mathematics, 4.01

Bachelor’S Degree, Mathematics, 3.98 Gpa