NLP and education
A chemistry-oriented chatbot prototype focused on narrow-domain question answering, corpus grounding, retrieval logic, and the practical application layer around NLP work.
The chemistry corpus contains 118 element records with element name, symbol, atomic number, and cleaned reference text. Source screenshots show question-answer examples such as element summaries, uses, and chemistry relevance prompts.
The broader NLP source folder also includes a sentence-transformer semantic chatbot over a separate stocks-and-bonds corpus. I use that as supporting evidence for retrieval-style thinking, while keeping Walter framed as the chemistry-domain project.
The notebooks use a practical retrieval pattern: preprocess the user question and corpus text, represent text with vector methods such as TF-IDF or sentence embeddings, rank candidate passages with cosine similarity, and return a grounded answer through chatbot response logic.
This makes the assistant easier to evaluate than an unconstrained chatbot because failures can be inspected at the corpus, query, ranking, and response layers.
This is best presented as an NLP proof of concept. The strongest evidence is qualitative: saved query-result examples, corpus coverage, and a retrieval pipeline that can be manually reviewed for relevance.
A more rigorous next version would add a labeled evaluation set, top-k retrieval review, relevance scoring, and failure categories for ambiguous or out-of-corpus questions.
