NLP Engineer
Text-To-SQL
Python, Postgesql, FlanT5, Fine Tuning, Prompt Engineering, FastAPI, Streamlit
Lifted valid SQL from ~38% to ~77% with ~83% execution accuracy across 11 cross-domain Spider databases.
LLMind
ChatGPT, DeepSeek, FinBert, Claude, Python, Prompt Engineering
FinBERT led on MSFT (R² = 0.97), Claude topped untuned models on AAPL/AMZN (R² = 0.88) — showing domain fine-tuning beats model size.