I have a growing collection of peer-reviewed papers that I need to turn into structured insight rather than long PDFs. The core objective is to build (or fine-tune) an artificial intelligence model that can read scientific articles, understand their content, and return usable analytics. In practical terms, the system should ingest batches of text-based research papers, then automatically extract key findings, highlight recurring themes, and provide concise summaries that I can export for further work. Because the project is about analysing data—specifically text data from scientific literature—I expect strong natural-language-processing skills with Python, Hugging Face Transformers, spaCy or similar libraries. Experience with science-specific language models such as SciBERT, BioBERT or GPT-based embeddings will be highly valued, as accuracy and domain nuance are critical. Deliverables • A working, documented pipeline (notebook or script) that accepts raw PDF or plain-text articles and outputs structured summaries, keyword/topic lists, and any relevant quantitative metrics (e.g., citation extraction, frequency counts). • A brief technical report explaining the model choice, preprocessing steps, and instructions for retraining or expanding the corpus. • Source code in a Git-enabled folder, with clear instructions for local setup (requirements.txt or environment.yml). Acceptance criteria 1. Given a test set of 20 unseen papers, the system must generate summaries that capture the main objective, methodology, and conclusions with at least 80 % ROUGE-1 recall when compared to human abstracts. 2. Key topics must align with domain-expert tags in at least 18 of the 20 papers. 3. All code runs on a fresh environment in under 10 minutes and requires no licensed dependencies beyond those named in the report. If you can turn dense scientific prose into actionable insight, I look forward to seeing your approach and a short outline of your preferred models and evaluation strategy.