WorkCompleteJanuary 2026personal

Nexus Graph

A tool that indexes academic papers and extracts their structure — authors, institutions, methodology, and materials — using an LLM.

The problem

Working through a literature review by hand is slow: cross-referencing authors and affiliations, and pulling out methodology and materials from each paper, is repetitive work that a language model can help structure.

Why it mattered

Turns a pile of PDFs into something you can actually query — who wrote what, with whom, using which methods and materials.

Topics

Built with

Architecture

The pipeline pulls in papers, then runs them through an LLM-driven extraction step that pulls out authors, institutional affiliations, methodology, and any materials or tools referenced. The results are cross-referenced into an indexed database so a paper, an author, or an institution can each be looked up on its own.

Engineering

The system architecture, indexing pipeline, and organisational modules are built in Python, with the OpenAI API doing the extraction work — turning unstructured paper text into structured records rather than leaving it as prose to be read one paper at a time.

Results

The tool works end to end: it categorises papers by field and topic, builds cross-referenced author/institution records, and structures methodology and materials information per paper. Precise before/after numbers aren’t published here yet — ask if you want the detail.

Lessons

Extraction quality depends heavily on how a paper is structured — papers with non-standard formatting or scanned figures need more handling than the straightforward cases. Widening database coverage and improving the extraction prompts are the two clearest next steps.