Supakit Boonsongprasert

Research Assistant @University of Maryland Baltimore County

Baltimore, MD, US

Designed and optimized a model pipeline to enhance the performance of narrative suggestionsystems, achieving a 60% increase in model accuracy and efficiency- Developed a narrative sequence generation model leveraging non-LLM baseline models(GPT-2 and SBERT) to compute log probability and similarity score, enabling structured storytelling or sequence prediction- Employed the subsequent LLM-based models: GPT-4.1, Gemini-2.5-pro-preview, Claude Sonnet-4, DeepSeek R1-0528, and o3, to produce narrative sequences from a variety of prompts with zero-shot, one-shot, and chain-of-thought (CoT) prompting strategies- Leveraged statistical techniques like Kendall Tau to evaluate the correlation between narrative sequences in the graph and the expected order of events. It quantifies how well the actual story structure aligns with an ideal or coherent structure.