Jake Kinney

- Achieved and maintained 90% annotation accuracy on dataset for state-of-the-art SAM 3 visual grounding model by performing regular quality audits, designing labeling metrics, and managing data engines - Evaluated and compared state-of-the-art generative AI large language models using a RLHF framework to improve model accuracy - Leveraged subject matter expertise to recommend optimization improvements for numerous annotation tasks, keeping models aligned to overall goals of the projects and resulting in improved output quality and user trust - Liaised with external vendors, implementing policy guideline updates, resolving disputes, and unblocking obstacles by escalating bugs and tooling issues with necessary documentation to engineering teams - Provided mentorship to new hires, accelerating their ramp-up and fostering a collaborative learning environment on the SAM team - Contracted full-time through Magnit

- Developed language resources such as context-free grammars and gazetteers for new features requested by customers - Analyzed accuracy reports to find systematic patterns of errors and investigate their sources in the data, resulting in accuracy improvements in customer test sets of over 10% - Corrected large amounts of annotated data using regular expressions to align lexical resources with evolving annotation specs - Led subproject on English NLU team, including triaging and delegating bug tickets when necessary and being the main dev for language resources for this project - Built and published English NLU datapacks, liaising with other teams to ensure quality and punctual delivery to customers - Contracted full-time through TEKsystems

- Designed and taught 3-week special relativity course to 100+ high school students in Prato, Italy

- Analyzed particle accelerator data with ICL's High Energy Physics Group as a contributor to the Large Hadron Collider (LHC)'s CMS experiment as one of 14 MIT students selected to participate in the competitive MIT-Imperial Exchange - Wrote deep learning neural network to predict particle mass in high-energy collisions and identify most important particle measurements

- Planned, organized, and executed team-building and engaging activities as a member of a 24-person counseling team - Supervised 85+ campers aged 12-18 and acted as a role model while guiding campers in personal growth

- Optimized jet tagging algorithms for the LHC's LHCb experiment - Improved efficiency of tagging processes using deep learning models

- Developed new pedagogy for teaching introductory physics at a college level with the education research group RELATE (Research in Learning, Assessing, and Tutoring Effectively) - Examined data collected from MIT classes in order to improve assessment techniques and teaching strategies

Coursework: Statistics, Probability & Random Variables, Fundamentals of Programming, Language Acquisition, Syntax, Semantics, Phonology, Constructed Languages, Real Analysis, Topology, Quantum Mechanics I-III, Statistical Mechanics, Experimental Physics, Differential Equations

Coursework: Computational Linguistics, Experimental Pragmatics, Quantitative Methods for Linguistics, Advanced Syntax, Linguistic Field Methods, Semantics, Phonology, Indigenous Languages of Latin America