AmeerAli Khan

• Authored and co-authored over 2 proposals, 3 papers, significantly contributing to the success of the team's services and application R&D efforts. • Fine-tuned Large Language Models (LLMs) for diverse tasks, including the LLAMA model, improving the ICD code classification accuracy by 3%, and the Falcon model for detecting software mentions in academic articles. • Contributed to securing approximately €50,000 in internal funding.

• Fine-tuned a T5 Language Model for Python method generation and code summarization tasks, improving the base model's performance and accuracy by 20%. • Trained a custom tokenizer on multiple programming languages, increasing the model's code understanding.

• Transitioned the existing Oracle DB to AWS Redshift through AWS Glue services. • Wrote PySpark scripts and constructed a Data Pipeline leveraging AWS Glue Jobs.

• Developed a predictive regression model using traditional ML algorithms to accurately estimate traders' profit or loss, significantly contributing to more informed decision- making processes. • Engineered a data pipeline utilizing AWS Glue and EMR with PySpark, for data preprocessing and feature engineer workflows, cutting down manual processing time by almost 100%.

• Developed NLP products at UnFound to address misinformation in journalism, compiling extensive news article datasets for analysis. • Collaborated with journalists to understand feature requirements, ensuring the development of NLP products that effectively address misinformation in journalism. • Designed and deployed product features and APIs on AWS for efficient data processing and access.

Thesis: Combined Preference-Based & Absolute Reward Signals for RLHF Fine-tuning Github Link: https://github.com/Ali1858/crs-rlhf • Developed a novel fine-tuning methodology, achieving a 66% win rate in evaluations over conventional approaches, enhancing model performance significantly. • Earned a Master's degree with a grade of 1.9, and achieved a 1.7 grade for the thesis, reflecting strong academic performance.