Role

• Developed an end-to-end financial question answering system using Large Language Models (LLMs) on FinQA dataset, improving end-to-end accuracy by 7% using hybrid embedding methods and advanced negative sampling.
• Designed a negative sampling method based on clustering, improving the RoBerta-Large based Retrieval model’s R@3 metric by 2.4% on test set, leading to SOTA results on FinQA.
• This work became the FinQAPT paper, Best Paper Runner-Up at ACM ICAIF 2024.
• Implemented Chain of Thought (COT), self consistency and self prompting methods with GPT-4, Llama2 and PaLM Large Language Models. <!– Heading 1 ======

Heading 2

Heading 3 ====== –>