literature-review-agent

Automating literature review verification requires fuzzy title matching and abstract-presence validation against Semantic Scholar, ensuring only deduplicated papers enter the citation pool. This process enforces a strict research cutoff to prevent prior-basis leakage and matches verified papers to structured related-work clusters.

Generating BibTeX files and LaTeX related work sections from verified sources involves deduplicating papers by paperId or DOI, producing a refs.bib file with canonical cite keys, and drafting template.tex using only the verified pool with a strict citation coverage gate.

Fuzzy title matching for Semantic Scholar citation verification uses python-Levenshtein similarity scoring to validate discovered web sources against authoritative database entries. It runs sequential 1 QPS throttled queries with caching to confirm abstract presence before allowing papers into the trusted citation pool.

Drafting arXiv paper introductions with a strict research cutoff is supported by enforcing temporal limits during discovery and verification. The system prevents prior-basis leakage by only integrating literature published before the specified cutoff into the Introduction and Related Work sections.

A 90% citation coverage constraint ensures guardrailed writing by forcing the generated Introduction and Related Work sections to cite almost exclusively from the verified, deduplicated literature pool. This prevents citation inflation and maintains rigorous academic standards.

python-Levenshtein is required to perform fuzzy title matching during Semantic Scholar verification and deduplication. Without this dependency, the sequential validation process cannot accurately confirm abstract presence or match discovered web papers to canonical database entries.