refchaser

Refchaser is an LLM-assisted tool that takes a scientific paper and builds structured context from its references.

Given a PDF, it:

• extracts the reference list
• parses each citation into structured metadata
• finds the referenced papers (via Semantic Scholar)
• pulls citation context from the main text
• generates short summaries of what each reference contributes

Reference parsing and summarisation use an LLM, retrieval and matching are deterministic.

The result is a per-reference “context view” of the paper: what each citation is, how it appears in the text, and why it is being used.

Why this exists

I'm too lazy to chase down all my references any more

What it does

For each reference, refchaser produces:

• parsed citation (title, authors, year, DOI)
• citation context (snippets from the main text)
• paper lookup (matched external record, if found)
• short summary (what the cited work is and why it’s used)

---

Quickstart

1. Setup environment

conda env create -f context-engine-env.yml
conda activate context-engine

export OPENAI_API_KEY=<<key>>
export S2_API_KEY=<<key>>

2. Run refchaser

python run_paper_context.py kalinich-et-al-2026.pdf

Example output (simplified)

{
  "reference": "O’Toole et al. 2021...",
  "found_in_main_text": true,
  "context_snippets": [
    "…used to assign viral lineages (3)…"
  ],
  "paper_lookup": {
    "title": "Assignment of epidemiological lineages...",
    "year": 2021
  },
  "summary": "Introduces the pangolin tool used for lineage assignment in viral genomics."
}