Final Year Project - Gamified Code Learning Platform with NLP
[2025]
This is a research project to improve the accuracy of agentic AIs in translating natural language to SQL statements, achieving top 10 on Spider 2.0-Snow evaluation benchmark with off-the-shelf SOTA models and agentic prompt framework (as of Apr 2025).
Developed agentic prompting framework “Tarantula” to iteratively handle errors resulting from complex Text-To-SQL tasks.
Improved agentic prompting framework based on ReAct, using methods such as grounding, RAG, prompt engineering.