Machine Learning Intern at A*STAR SIMTECH
Developed time series forecasting models on NHITS and TFT architectures, improved accuracy by 30% Finetuned TimeGPT and Chronos LLMs for feature reduction, decreasing computation time by 40%
Developed time series forecasting models on NHITS and TFT architectures, improved accuracy by 30% Finetuned TimeGPT and Chronos LLMs for feature reduction, decreasing computation time by 40%
Designed a robust CV extraction pipeline with Reducto, reducing customer intake time by 60% Constructed an end-to-end automated report generation tool for pharmaceutical ingredient analysis
Extended recurrent convolutional neural networks (RCNN) for surface code quantum error correction Tuned RCNN model to achieve equivalent performance with less than 25% of original parameter count Quantized TensorFlow dense models with hls4ml for use in dedicated FPGAs, reducing load by 50%
Utilized retrieval augmented generation with agentic LLMs to generate 200+ multiple choice Q&As Automated Q&A generation and verification pipeline with over 95% efficacy for the Aurora GPT Developed a tutorial on agent-based LLMs using AutoGen and LangChain for over 50+ instructees
Helped develop a Python program that collected internet speeds throughout Hyde Park using Ookla’s API and projected them onto a map through GeoPandas to see differences in internet accessibility throughout the neighborhood
Intro The goal of this project was to collect internet speeds in counties throughout California and explore the reasoning behind the differing county speeds. (Github Link) Findings Pictured below are the top 20 counties by average download speed (Mbps). Contrary to what many may think, only a few counties, like San Mateo, Contra Costa, and Alameda, ranked in the top 20 for both average internet speed and median household income (Census Bureau), but more than half of the counties with top 20 internet speeds did not intersect that top 20 income list....