Leveraged GPT-4 via OpenAI API to process dialogue from online interview videos, extracting key information such as skills and project highlights; conducted comprehensive data cleaning and preprocessing, utilizing Git for version control, leading to a 20% improvement in candidate matching efficiency

Fine-tuned prompt generation for dataset creation, employing ThreadPoolExecutor in Python to parallelize extensive question prompt lists, optimizing resource allocation and reducing computation time by 15%.

Applied SentenceTransformer to encode textual data, developing similarity comparison using cosine similarity to assess response consistency across multiple instances, providing more reliable evaluations of candidate responses

Conducted feature engineering by generating sentiment scores and keyword frequencies, enhancing the response datasets and improving the accuracy of insights derived from large language models (LLMs)