NLP/LLMs

Program Synthesis via LLMs/Uncertainty Estimation

Program Synthesis via LLMs/Uncertainty Estimation

• Designed and implemented a custom fully interactive data visualization tool for a research study• The requested data visualization was novel, with no standard implementations• Worked closely with client to go through multiple designs and rendering strategies, including R / ggplot2 and Python / Matplotlib before eventually settling on JavaScript / d3 / HTML Canvas• Tackled performance issues caused due to high number of recalculations occurring due to user interactivity using several strategies, such as creating a custom cached data store which would minimize update operations as much as possible and moving performance critical visualization elements from a d3 to a HTML Canvas implementation which directly manipulated raw pixel values• The optimizations lead to a significant performance boost, roughly 50%, resulting in a smooth user experience• The development of this tool was highly customer focused, with me working very closely and in constant communication with the research professor to adapt to the changing requirements of the project

• Created a data processing tool for an ongoing research study• The tool used a Python / Flask backend for processing data, and a JavaScript / React frontend• The processing portion synchronized time series data from different sources with different polling frequencies• It performed basic data cleaning as well as a set of frequently used data transformations as specified by the user• Originally, the processing tool had a Tkinter based GUI which was later discarded to provide an all in one tool that included the data visualization tool

• Designed and developed a prototype for a highly scalable streaming data auditing service to be used for firewall risk assessment• Created a RESTful API to provide access to data stored in the data warehouse• The prototype had two implementations– 1) traditional server based and 2) serverless architecture• The two implementations were used by the team for comparative analysis to guide future product development • Harnessed several AWS cloud technologies, including Kinesis Firehose, S3, Redshift, Lambda and API Gateway• Gained experience with test driven development, mock testing, dependency injection frameworks, agile development and worked closely with a team of 5 developers

• Wrote a tool to classify handwritten digits using Python and optimized it using a custom matrix library created in C++/CUDA• Placed 6/250 at WSU Hackathon 2015 for a text based assistant, along the lines of Siri• Designed and implemented the artificial intelligence for a bullet hell game for a class project