Using LLM for Research Purposes: A Guide

Description and use

An automated Python pipeline uses LLMs to screen, extract, and synthesize literature reviews, reducing manual effort.

Problem

Researchers need to survey large literature without reading every paper in full.

Outcome

The script automatically screens abstracts, extracts structured findings, and produces a concise, markdown literature review, reducing manual effort and ensuring claims are grounded in source text.

Workflow

• Initialize OpenAI-compatible client pointing at Oxlo.ai.
• Create a JSON corpus of paper abstracts.
• Define a system prompt that enforces no hallucination and structured output.
• Screen each abstract for relevance using a low‑temperature LLM and score 0‑10.
• Extract structured findings from papers scoring ≥5, returning JSON with key_finding, method, limitation, confidence.
• Synthesize a markdown literature review from aggregated findings using a reasoning‑capable LLM.
• Print the final report.
• Optionally extend to PDF parsing and larger models for full‑text papers.

Evidence

We read each abstract and ask the model whether it addresses our research question.

Source

Extract structured findings from this paper. Return valid JSON: { "key_finding": "string", "method": "string", "limitation": "string", "confidence": "high|medium|low" }

Source

Write a short literature review with these sections: 1. Executive Summary (2-3 sentences) 2. Thematic Synthesis ...

Source