Academic Deep Search & QA
A local research assistant that searches Semantic Scholar for a topic, filters and summarizes the papers it finds, and answers questions grounded only in that retrieved set.
Academic Deep Search & QA is a local implementation of a Google-style deep search assistant for academic literature. Given a topic, it bulk-searches Semantic Scholar, filters results by publisher or journal match and by whether an abstract exists, then uses a language model to judge relevance before building a summary tree over the kept papers by clustering and summarizing chunks repeatedly up to one root.
Answers to follow-up questions are retrieved from that tree through FAISS and grounded only in the retrieved nodes. Chat, summarization, and embeddings all go through a local llama.cpp server over its OpenAI-compatible API, usually run as two separate processes, one for chat and one for embeddings. Exports include an academic-style PDF report, a diagram of the pipeline, and a bibliography of both kept and discarded papers. The reusable logic lives in an installable Python package; Streamlit provides the interface.