asma 0.1.0

Automated System for Mining Articles (asma)

asma — Automated System for Mining Articles

asma is a modular, extensible Python library designed to automate the ingestion, parsing, and structured extraction of scientific research articles from PDFs using NCBI PMC APIs and local Large Language Models (LLMs).

Features

• Automated DOI Extraction & Validation: Lazy-loads PDF processing utilities to scan and validate DOIs via Crossref.
• NCBI PMC Ingestion: Resolves DOIs to PMCIDs and PubMed IDs with robust exponential backoff.
• Dual-Purpose BioC Parsing:
  • LLM-optimized Markdown: Strips references and serializes tables to raw CSV layout to conserve tokens and improve accuracy.
  • Human-optimized Markdown: Preserves references and builds clean Markdown tables for easy reading.
• In-Context Prompt Engineering: Decouples instruction logic from schema definition fields to support dynamic prompting.
• LLM Provider Agnostic: Interface-driven (LLMProvider) to easily swap between LM Studio, Ollama, OpenAI, or other backends.
• Automated Validation: Evaluate extraction outputs against ground-truth files and generate markdown report cards.

---

Repository Structure

├── src/
│   └── asma/              # Main library source code
│       ├── core/          # Markdown parsing and Evaluation engine
│       ├── providers/     # Crossref resolvers, PMC fetchers, LM Studio client
│       ├── utils/         # PDF helpers, XML parsers, text utils
│       └── config.py      # Prompt templates and default schemas
├── tests/                 # Unit test suite
├── run_pipeline.py        # End-to-end command-line orchestrator
├── pipeline.ipynb         # Interactive Jupyter demo notebook
├── pyproject.toml         # Package definition (PEP-621)
└── asma_documentation.md  # Detailed SDK reference & developer guide

---

Quick Start

1. Installation

Install the package in editable mode:

pip install -e .

To enable local PDF DOI extraction, install the PDF support extras (installs PyMuPDF):

pip install asma[pdf]

2. Run the Orchestrator

To run the pipeline end-to-end (requires a local model server loaded on LM Studio):

python run_pipeline.py 36374021

3. Detailed Documentation

For a comprehensive guide covering custom schemas, extending providers (like Ollama), streaming controls, and the testing framework, read the Developer Reference Guide.