cdQA

Why we included this project

cdQA packages the retriever/reader pattern that still underlies many knowledge-base assistants: a sparse retriever pulls candidate passages from your own documents, and a BERT or DistilBERT reader extracts the answer. The code is compact and readable, and the repo includes converters that turn PDFs and markdown into the pandas dataframe the pipeline expects, plus helpers to fetch SQuAD-fine-tuned readers and evaluate predictions against annotated sets. The project is unmaintained, and the authors say so plainly, pointing to Haystack as the maintained alternative. That makes it a learning reference rather than something to build on: a worked example of how retrieval and reading comprehension fit together, with a Medium article and notebooks that walk through the design decisions.