• Submitted multiple papers to IEEE, WACV and ECAI, prestigious Computer Vision and AI research conferences• 6% performance increase via contactless biosignals extracted from computer vision techniques for Engagement Video QA to base finetuning• 6% performance increase using Direct preference optimization to generalize to both in‑sample and out‑of‑sample data to base finetuning• Achieved 97% F1 precision recall on student engagement dataset using Engagement Level prompt finetuning for Visual QA task• Self generated preferred/rejected response pairs without human annotation for Direct preference Optimization

• Increased 200,000 SOI data samples from 20 different BDCs to model SEC registered BDCs fiscal product categorization & valuation• Nullified need to post process data using OpenAIs gpt4o‑mini, reducing data pipeline staging processes by 50%• Prototyped and implemented staged Data models for ELT Data pipeline and Database for single source of truth mapping• Containerized data extraction and data cleaning programs for future use and cross platform portability

• Abstract away need for traditional data infrastructure using AI as a service(AIaaS) paradigm while maintaining client data security and minimal technological overhead• Innovate with Multi‑Agent GenAI BIM Vector and Graph based Retrieval Augmented Generation POCs for client demonstration • Landed partnerships with PennMedicine and several other Hospital Authorities in PA, NJ and NY through built Compliance Front end• Assisted in PA,NJ,NY Port Authorities Data analytics development

• Led Computer Vision research at National Taiwan University to enhance Googles testing suite for detecting UI defects on their Pixel Mobile• Enabled further Google level 2 sponsorship at National Taiwan University through further research in to Multi‑Modal Computer Vision• 43% benchmark to 98% precision‑recall enhancement for UI Flicker Detection using CNN‑Transformers model• 10x increase in data augmented samples, 180 to 1991 videos snippets• 20x reduction in model training time through appropriate initialization of weights and bias, 1000 epochs to just 50 epoch

• Gave qualitative insights to CPC Taiwan Oil to decide whether having a new oil rig near the coast of Taoyuan would damage its local ecology• Utilized Structured Equational Models to conduct Conceptual causal analysis of latent regressor variables and observed response variables

Research Associate for research summaries • "Do you need a blockchain?", • "Towards private on-chain algorithmic trading" • "Machine Learning Guided Cross-Contract Fuzzing" for Smart Contract Research Forum Research Associate for SCRF Engagement team; • Seed insight conversions from SCRF internal discord onto forum • Seed insight conversion for research summaries, discussion posts, research pulses, podcasts, and community calls • Be a moderator for SRCF engagement for the APAC region

The SCRF Research Mentorship Program is a scholarly program to enhance your understanding of research focusing on researching and developing research summaries and articles. Participants will also receive training on the SCRF process and policies. There will be an emphasis on researching and developing research summaries/articles to strengthen the SCRF community while also providing valuable knowledge about the blockchain ecosystem and technologies.

• End to end Recommendation System integrating layer 2 blockchain protocols with layer 3 Web UI interfaces, Blockchain with Data Science• Design ERM and Database administration policies for highschool students tagged to a recommended university’s departments.• Decoded combinations of big‑5, utf‑8, and Cp1252 data encodings to unify scrapped data from different languages

• Research inverting confidence intervals for systems of multivariate higher order linear combinations.• Project application to develop algorithm that simultaneously finds multiple solutions for complex calibration systems.(Mar Rover HeatShield)• Bootstrap Confidence Intervals for Xgboost Stocks percent change responses with an root mean squared error of 0.6, near perfect fit.• Spotify predictive monthly artists key performance analysis via collinearity analysis, Variance inflation factor analysis,PCA regression.

Meeting individuals and professionals from all walks of life exchanging my computer science knowledge with their personal and professional insights

• Reduced fps stitching speed of 2x 3840X2160 pixels, 170 degree fisheye cameras from 6 seconds to 0.2 seconds on the Nvidia Jetson Xavier• EstimateTransform only once for real time composePanorama with multi‑threading, camera is stationary but capture is 360 degrees• Reduced 0.2 to 0.01 fps using CUDA code integration of opencv 2.4 image blending into opencv 4.5 stitching during composePanorama• 6x reduction in opencv cuda/cudnn binary compilation time by generating CPack binary deb files for single Opencv compilation (6 to 1 hours)• Implemented real time 3d video to 2d image projection homographies of player field positions• Developed automated fisheye to panaromic view camera callibration