Student Q&A Assistant

The Student Q&A Assistant was built to take the load off student administrative enquiries — the repeated questions that consume staff time and leave students waiting.

The pipeline is deliberately classical rather than trendy: text cleaning and tokenisation with NLTK, feature extraction using TF-IDF alongside Word2Vec and FastText embeddings, and intent classification with Naive Bayes and KNN models.

The result is measurable: 78% of queries were resolved autonomously, without a human in the loop, while low-confidence questions still route to a person rather than guessing.

The product wraps this in a Flask API with a React chat interface, so the model stays a service and the interface stays free to evolve.

• Built an AI chatbot with Flask and React for student administrative queries.
• Achieved 78% autonomous resolution without human intervention.
• Implemented NLP techniques with NLTK: TF-IDF, Word2Vec and FastText.
• Trained and compared Naive Bayes and KNN classification models.
• Designed a confidence-aware escalation path so uncertain answers reach a human.
• Kept the model behind a Flask API so the interface and model can evolve independently.

Conversation flow

Students ask in natural language. The assistant classifies intent, retrieves the relevant administrative answer and only escalates to a person when confidence drops.

NLP pipeline

Text is cleaned, tokenised and vectorised with TF-IDF, Word2Vec and FastText embeddings, then classified with Naive Bayes and KNN to pick the right answer.