various llm related tasks, received return offer to come back for third summer

For personal project, envisioned and developed:-proprietary reinforcement learning-based recommender system (in use by Morgan Stanley investment management)-reinforcement learning assisted tool for sequential feature explainability from intractable deep learning models (in use by Morgan Stanley investment management)-novel process for efficiently converting arbitrarily large multivariate sequential time series data following certain constraints to finite stochastic control processes (implicit in above projects)Received 2-week extension, return offer Personal project taken over for continuation by full-time researchers after end of internship

-Worked on calibrating low-cost particulate matter sensors in Kolkata, India, and predicting the burden of pollution across the entire city from the calibrated sensor data.-Fixed existing data errors in pollution data for ColumbiaClimate School and EPA India pollution data- Built models utilizing sparse co-located government-run pollution sensor information and corresponding weather API data to calibrate low-cost sensors.- Built heat-map pollution model from scratch usinggaussian process regression

Worked on curriculum for Harvard undergraduate data scienceclass Gov50- In R, Rstudio: authored in-use textbook sections on mapping and Public Use Microdata Sample (PUMS) data- Wrote in-use tutorials and homework questions on datawrangling and visualization in R

Designed and taught science lessons and tours to visitors at the museum- Facilitated touch carts with artifacts and science experiments - Reaccepted to the program for a second year before its discontinuation due to the pandemic