Enhancing customer-facing offers (assortment, availability, pricing) by developing intelligence across inventory, selection, and pricing within the New Vertical Merchant Offer Platform Team

Led data ingestion for MultiSource Selection, synthesizing signals from merchants, POS, suppliers, and Dasher to infer accurate store-level on-shelf inventory—a challenge traditionally unsolved in retail; scaled ingestion of 12B+ daily items via Cadence workflows and Kafka consumers

Devised and rolled out an experimentation framework for item-level pricing and inflation strategies in close collaboration with strategy and operations teams, enabling daily testing of 20M+ items and driving 1.4%–6% GOV lift for major merchants (Target, Schnucks, Home Depot)

Identified and resolved gaps in item pricing logic, preventing $2M annual GMV loss and correcting 1.6M+ mispriced items/year

Developing AI design tools, focusing on end-to-end machine learning applications that empower 5000+ users on Pietra Platform

Finetuning Stable Diffusion XL on Google Cloud with multi-GPU to enhance the realism of fabrics on clothing design images.

Engineering an automated Tiktok Ads generation pipeline with GPT4o, Google Cloud text-to-speech API, and lip sync model

Enhancing SEO ranking of AI design tools on 50+ major recommendation sites, boosting Google search ranking from 30+ to top 3

Selected as top 0.5% of applicants, contributing to the libsecp256k1 library, the highest-quality public cryptography library on the secp256k1 curve used in Bitcoin

Developing an adaptor signature module for BIP 340 signatures (Schnorr Signature) to enable atomic swap without Bitcoin Script and PTLC in the Lightning network, which improves privacy and transaction efficiency

Designed an optimized 4-step scheme of the Schnorr adaptor signature, including presigning of an adaptor signature, adaptor point extraction, adaptation of an adaptor signature into a Schnorr signature, and adaptor extraction

Replaced the original verification step in digital signatures with extracting adaptor point which resulted in space-saving of over 25%

Created a python specification for validation, and built unit test cases and test vectors in each API development cycle

Initiated weekly code reviews with the open-source community, resolving 100+ PR reviews

Worked with Professor John Kender in High Level Vision Lab, and contributed to an NSF IIS-funded project on cross-cultural multimedia analysis, analyzing differences in scene patterns between the U.S. and China insurance commercials

Employed YouTube API and web scraping techniques to build a dataset of 100+ commercials and designed a preprocessing pipeline using FFmpeg and Vision Transformer for video encoding, scene segmentation, and scene labeling

Utilized an 85%-accuracy decision tree model for feature extraction to encode commercials into scene sequences

Analyzed scene sequences with n-gram and Markov Model, providing insights to cultural-based commercial advertising strategies

Enhanced the performance and efficiency of the binarization step in OCR (Optical Character Recognition), and devised a novel corner detection algorithm for an industrial quality-checking system

Developed an adaptive threshold algorithm using OpenCV and C++ to binarize unevenly lighted images of ingredient list on food packaging in OCR, reducing the worst-case time complexity from O(w^2mn) to O(mn) compared to traditional approaches

Implemented a lightweight algorithm utilizing binarization to locate the left corner of a prismatic hole in images, streamlining the process from 7 steps to just 3 compared to the original Harris corner detector algorithm