TIGER-AI-Lab/ScholarCopilot

Integrates a unified retrieval-generation architecture that dynamically switches between text generation and citation prediction, using learned citation patterns to identify insertion points and retrieving relevant papers via HNSW-indexed embeddings from arXiv corpus. Supports fine-tuning on custom academic datasets and enables corpus updates with new papers through metadata processing and embedding generation pipelines. Provides inference capabilities for next-3-sentence suggestions, full-section auto-completion, and one-click bibtex citation insertion.

251 stars. No commits in the last 6 months.