Text 2 SQL AI

As a Computer Science graduate, I developed Text2SQLAI, an innovative web application that translates natural language queries into accurate SQL statements, streamlining database interactions for non-technical users. Leveraging a fine-tuned T5 model (a transformer-based NLP model), I designed a system that processes queries like “Count the number of sales” or “Show products with name containing phone” and generates corresponding SQL, such as SELECT COUNT(*) FROM sales or SELECT * FROM products WHERE name LIKE '%phone%'. I implemented a robust rule-based fallback in Python to handle cases where the T5 model underperforms, ensuring reliability across 11 test queries. The project involved creating a Flask backend, a responsive Vercel-hosted frontend, and integrating spaCy for entity extraction, all hosted in a local development environment. The development process required overcoming challenges like debugging a TypeError in the SQL generator and resolving a “Not Found” error for the frontend. I iteratively refined sql_generator.py and nlp_processor.py to achieve 10/11 correct queries, with ongoing work to perfect the final query (“Get the id and product from sales where amount equals 1000”). The project is version-controlled on GitHub (https://github.com/LiamMedenica/Text2SQLAI), showcasing my ability to manage a full-stack AI project. Text2SQLAI demonstrates my skills in NLP, web development, and problem-solving, with plans to enhance it with JOIN support and deploy it to Vercel after further retraining.