- Launched international 501(c)(3) computational linguistics outreach nonprofit (http://linghacks.tech) partnered with CSforALL National Consortium & Major League Hacking- Developed & taught 25+ curriculum modules for computational linguistics workshops (Workshops on Words pilot series, partner workshops @Girls Code Month, Girls Code Data) & clubs across 21 cities in US (CA, CO, NC, HI, TX, PA, NY), Canada, India, Morocco, China- Founded & ran world's first high school computational linguistics conference & hackathon (2x), featuring speakers from Google, Stanford, Salesforce, AIBrain, USC, SDL, General Assembly, Cal Poly SLO- Created & produce content for educational blog, email newsletter, & resource page featuring technical resource compilations & tutorials on programming, machine learning, natural language processing (linghacks.tech/blog, linghacks.tech/resources)- Raised $17.5k+ in monetary sponsorships (incl. 2x inaugural AI4ALL Impact Grant) & $55k+ in in-kind sponsorships, managed executive team of 20+, gained sponsorships/partnerships from 40+ organizations, recruited 80+ volunteers/mentors/judges/speakers/organizers, impacted 700+ students (participants exited LingCon/LingHacks I 39% more likely to pursue NLP careers on average)- Advised, recruited event staff, spoke on panels, taught workshops, judged submissions for LingHacks III, IV, & V

LLM safety 🤫

- Created novel benchmark dataset for bias-related harms of model editing- Investigated demographic biases & misinformation in edited large language models- Annotated labmates' data & gave feedback on their works in progress

- Co-produced DanceFest & Science Center Plaza Stage performances for Harvard’s annual public arts showcase- Communicated with 15+ dance groups & technical staff to assemble performance fair program & gather stage diagrams, tech needs, & music files- Publicized performance fair through maintaining performer roster by department & delivering postcards- Stage-managed DanceFest as backstage point of contact, supervised t-shirt distribution, ran tech rehearsals, answered patron questions, assembled puzzle mats for children- Created show order & schedule, shared information with emcees, managed volunteers- Choreographed & performed original character dance solo in DanceFest

- Migrated & maintained website deployment from Netlify to Hugo + GitHub Actions, saving $19/month- Co-recruited & oversaw team of 15 TFs- Planned & co-led pre-semester push, staff meetings, & heads meetings- Answered emails, resolved students' technical & administrative issues with GitHub & Gradescope, regraded problem sets, released lab solutions- Implemented & maintained public health communication & support system (COVID exposure notifications, PPE provision, remote office hour/section/lab infrastructure for isolating students)- Tracked attendance (helped with onsite lab checkin, sent & filed bug report for end-of-semester attendance reports, monitored excused absences)- Created & maintained Ed best practices, homework FAQ sheet, data structure explainers, staff to-do & FAQ database- Staffed labs, taught Code Review, graded assignments, proctored remote & in-person midterms, answered 1100+ Ed questions, fixed 7 autograder bugs, styled lab & homework distribution code, created text expander files to facilitate narrative feedback in homework grading

- Greeted patrons, ripped & counted ticket stubs, distributed programs & masks, answered patrons' questions for orchestral & choral concerts, dance installations, movie screenings, & guest speaker events- Communicated with house manager & stage manager on radio- Loaded & unloaded percussion, string instruments, chairs, stools, & music stands & placed chains for lift runs during set changeovers- Moved benches, chairs, music stands, stanchions, & ropes onto/off of racks & carts

- Prototyped computational visualizations of relationships between legal documents to help social science researchers gain faster, more big-picture understandings of legal primary sources- Built OverText_{history}, a novel interface that enables historians to make side-by-side sentence-level comparisons of keyword occurrences within groups of primary sources, grouped into tables by keywords & top neighbors, sorted within tables by clusters of semantic similarity- Wrote senior thesis on OverText_{history} design & application to historical case study on Ottoman Sephardi women's experiences of wars from 1908 - 1945

- Built centralized Server Resource Management service that calculated memory & CPU usage stats for all processes on a host- Improved Tableau Server resource monitoring efficiency from O(N^2) to O(N)- Increased reliability of resource usage information communication to client processes through TCP & enabled Tableau to monitor resource usage of non-C++ processes- Tools: Java, C++

- Staffed labs, held office hours, graded assignments, answered 2000+ Ed questions, taught weekly sections, contributed fixes to textbook & labs, created & ran study buddy matching program

- Staffed labs, graded homework, held weekly office hours, answered questions on Ed, checked students in for lab, taught section on Hidden Markov Models/Viterbi Algorithm

- Built WatSense, a C# application to enable users to prioritize & triage Windows bugs by driver & operating system- Wrote functionality to rank a bug on its driver & each OS, view top n bugs on a driver & by OS, view patch-Tuesday-ly bug report trends, view Visual Studio Online work items for a bug, & view earliest + total hits for a bug- Saved Windows Servicing & Delivery's IPCOMP team > 1 hour per bug fix per person- Tools: Microsoft Watson API, Visual Studio Online API

- Taught weekly sections on software design, functional/modular/imperative/lazy/object-oriented programming, & semantics using OCaml- Held weekly office hours, graded homework/exams/projects, staffed labs, beta-tested course materials, wrote guides for grading/pair programming

- Developed recommendations & delivered educational equity workshops to graduate teaching fellows in math, anthropology, & engineering/applied science departments- Designed listening sessions to incorporate stories & insights from members of student affinity groups into educational equity/accessibility workshops

- Created 212 mathematical concept visualizations in R for Math 23b (linear algebra, real analysis, multivariable calculus II)

- Developed original course materials & taught sections for world's first course combining abstract math (discrete math, linear & abstract algebra, topology, graph theory) with R Shiny web development & data analysis- Helped students debug code & understand math in weekly office hours- Graded homework & projects, wrote comprehensive exam review guide & practice final

- Tutored students in introductory computer science & math (linear algebra, real analysis, multivariable calculus)- Helped students review for exams & develop strategies for homework

- Led 25-student section- Facilitated interactive seminar presentations & group problem-solving workshops

- Wrote real analysis workshop problem solutions using LaTeX- Created 133 mathematical concept visualizations on linear algebra, multivariable calculus, real analysis, finite topology, & category theory using R- Developed & deployed 3 R Shiny web apps on modular arithmetic & vector transformations used in Math 23a (Linear Algebra & Real Analysis I) & Math 152 (Discrete Mathematics)- Co-taught R Shiny tutorial to students in 3 Harvard Summer School courses

- Collected historical data on policing & prison spending in Western European countries for project investigating American exceptionalism in crime & punishment- Used R to develop automated data source merging pipeline, read archival data, & generate visualizations of areas in which data was missing

- Co-developed & tested categorical modularity, a novel low-resource intrinsic metric for word embedding evaluation, on 29 languages, 3 embedding models (FastText, MUSE, subs2vec), & 3 downstream tasks (sentiment analysis, bilingual lexicon induction, word similarity)- Tools: Python, scikit-learn, NLTK, FastText, scipy

- Worked in Analytics & Machine Intelligence group on DARPA Active Interpretation of Disparate Alternatives (AIDA) program, focusing on named entity recognition (NER) for COVID-19 narrative-focused news articles- Created & annotated original news NER dataset with 1.7k+ sentence samples & 5k+ entities- Created classifiers & binary extractors on CORD-NER, NCBI-Disease, & news data- Added expression matches for biomedical entities to AIDA pipeline- Tools: BioBERT, Hugging Face, Perl, Python, scikit-learn

- Built recommendation engine to generate image suggestions based on news article content- Created backend service & front-end user interface for Bloomberg Terminal- Deployed version 1.0 to beta, completed development for version 1.1- Tools: Python, JavaScript, XML

- Created multi-label machine learning algorithm to detect & prevent cyberbullying- Presented research on "AI & The Information Sphere" panel at Stanford x UN OHCHR "Future of Human-Centered AI" conference- Tools: Python, NLTK, pandas, NumPy, scikit-learn, Keras, Excel

- Researched policy topics including bilingual education & personal data acquisition for proposal to Congresswoman Eshoo- Assembled final board reports on "Maximizing Opportunity" & "Sustainable Growth," both housed in Library of Congress

- Research intern at Kleiner-funded MIT Media Lab spin-off creating Emotion AI- Worked on multimodal (face + speech) frustration detection using deep neural networks- Empirically analyzed data labelled with prototypical emotions to generate "frustration," "anger," & "neutral" annotations- Built GIPHY web scraper that added 3600+ pieces of data to computer vision pipeline- Tools: Python, NumPy, Pandas, Keras, Excel, SQL, GIPHY Python Client