Research projects include： * Extracting information from documents such as Tax-Form, Employment Letters, etc. Responsible for customizing BERT NER labels, tokens label alignment, training data pre-processing, and fine-tuning BERT NER. Improved NER performance by 15% to an accuracy of 99.075%, F1 score of 99.147%.* Experimented on various FAISS indices for higher accuracy query document matching, including inverted file index, HNSW, LSH, and product quantization. Fine-tuned… Show more Research projects include： * Extracting information from documents such as Tax-Form, Employment Letters, etc. Responsible for customizing BERT NER labels, tokens label alignment, training data pre-processing, and fine-tuning BERT NER. Improved NER performance by 15% to an accuracy of 99.075%, F1 score of 99.147%.* Experimented on various FAISS indices for higher accuracy query document matching, including inverted file index, HNSW, LSH, and product quantization. Fine-tuned, parameter-tuned BERT Question Answering Models. Attained accuracy of 96.619%, F1 Score of 97.605% (Haystack, HuggingFace, BERT, FAISS)* Built a hybrid model for comparing the semantic differences between documents, through leveraging statistical methods and string metrics such as Levenshtein difference, TF-IDF, and a mixture model comprising of Sentence-Transformers and Sentiment Analysis, for the purpose of measuring contextualized semantic similarity at a sentence level. (C++, HuggingFace, Python, Sci-Kit Learn, NLTK, PyTorch).* Established a data pipeline for training, fine-tuning, and model validation, streamlining, standardizing, and automating data pre-processing, transformation, and curation procedures. (Datasets, PyTorch)* Built a synthetic data generator using custom templates that create simulated data by sampling from various statistical distributions with well-defined ontologies. (Python) Show less

* Designed trading strategies and operated on trading systems (market making, arbitrage, DeFi Arbitrage). Oversaw the development of strategies and systems (e.g.: Design meetups, UAT Tests, Code Reviews, etc.).* Conducted data analysis through programming, and model building, i.e.: identifying market opportunities, backtest strategies, risk management, post-trade analysis. (Python, Numpy, Pandas, Impala, Hive, SQL, Excel)

Weather Modelling - Data Assimilation of Ozone and Nitrogen Dioxide - Automated Data Acquisition (Satellite Data) - Development of Global Forecast System (GFS) modules - Data Cleaning - Model Fitting