llama-index-readers-builtsimple

LlamaIndex readers for Built-Simple research APIs, providing semantic search over scientific literature.

Features

• PubMed Reader - 4.5M+ biomedical articles with hybrid semantic/keyword search
• ArXiv Reader - 2.7M+ preprints in physics, math, CS, and ML
• Wikipedia Reader - Semantic search over Wikipedia articles
• No API key required - Free tier available for all endpoints
• Rich metadata - Full citation info for all documents

What Data is Included

PubMed Reader

Each document contains:

• Text: Title + abstract (default) OR full article text (with include_full_text=True)
• Metadata:
  • pmid - PubMed ID (e.g., "31041627")
  • title - Full article title
  • journal - Publication journal name
  • year - Publication year
  • doi - DOI identifier
  • doi_url - Direct DOI link
  • url - Link to PubMed page
  • has_full_text - Boolean indicating if full text was fetched
  • full_text_length - Character count of full text (when available)

🔥 FULL TEXT AVAILABLE! Unlike most research APIs that only provide abstracts, Built-Simple has full article text for millions of papers:

# Get full article text (15K-70K chars per article)
reader = BuiltSimplePubMedReader(include_full_text=True)
docs = reader.load_data("cancer immunotherapy", limit=5)

for doc in docs:
    print(f"Full text length: {len(doc.text)} chars")  # ~15,000-70,000 chars!

ArXiv Reader

Each document contains:

• Text: Title + authors + full abstract
• Metadata:
  • arxiv_id - ArXiv identifier (e.g., "2301.12345" or "cs/0308031")
  • title - Paper title
  • authors - Author names
  • year - Publication year
  • url - Link to ArXiv abstract page
  • pdf_url - Direct PDF download link
  • similarity_score - Semantic relevance score (0-1)

Note: Full paper PDFs are NOT downloaded—only abstracts. Use pdf_url to fetch the full PDF if needed.

Wikipedia Reader

Each document contains:

• Text: Article title + summary/intro section
• Metadata:
  • title - Article title
  • url - Link to Wikipedia page

Note: Only article summaries, not full articles.

Installation

pip install llama-index-readers-builtsimple

Quick Start

Basic Usage

from llama_index.readers.builtsimple import (
    BuiltSimplePubMedReader,
    BuiltSimpleArxivReader,
)

# Search PubMed for medical literature
pubmed_reader = BuiltSimplePubMedReader()
pubmed_docs = pubmed_reader.load_data("CRISPR gene therapy", limit=10)

for doc in pubmed_docs:
    print(f"Title: {doc.metadata['title']}")
    print(f"Journal: {doc.metadata['journal']}")
    print(f"Year: {doc.metadata['pub_year']}")
    print(f"URL: {doc.metadata['url']}\n")

# Search ArXiv for ML papers
arxiv_reader = BuiltSimpleArxivReader()
arxiv_docs = arxiv_reader.load_data("transformer architecture attention", limit=10)

for doc in arxiv_docs:
    print(f"Title: {doc.metadata['title']}")
    print(f"Authors: {doc.metadata['authors']}")
    print(f"ArXiv ID: {doc.metadata['arxiv_id']}\n")

Build a RAG Index

from llama_index.core import VectorStoreIndex
from llama_index.readers.builtsimple import BuiltSimplePubMedReader

# Load documents
reader = BuiltSimplePubMedReader()
documents = reader.load_data("immunotherapy cancer treatment", limit=20)

# Build index
index = VectorStoreIndex.from_documents(documents)

# Query
query_engine = index.as_query_engine()
response = query_engine.query("What are the side effects of CAR-T therapy?")
print(response)

Combine Multiple Sources

from llama_index.core import VectorStoreIndex
from llama_index.readers.builtsimple import (
    BuiltSimplePubMedReader,
    BuiltSimpleArxivReader,
)

# Load from multiple sources
pubmed = BuiltSimplePubMedReader()
arxiv = BuiltSimpleArxivReader()

# Combine documents
documents = []
documents.extend(pubmed.load_data("drug discovery machine learning", limit=10))
documents.extend(arxiv.load_data("drug discovery deep learning", limit=10))

# Build unified index
index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine()

response = query_engine.query(
    "How is machine learning being used for drug discovery?"
)
print(response)

API Reference

BuiltSimplePubMedReader

BuiltSimplePubMedReader(
    api_key: Optional[str] = None,  # Optional for higher rate limits
    timeout: int = 30,
)

def load_data(
    query: str,
    limit: int = 10,
) -> List[Document]

Document Metadata:

• source: "builtsimple-pubmed"
• pmid: PubMed ID
• title: Paper title
• journal: Journal name
• pub_year: Publication year
• doi: DOI identifier
• url: Link to PubMed

BuiltSimpleArxivReader

BuiltSimpleArxivReader(
    api_key: Optional[str] = None,
    timeout: int = 30,
)

def load_data(
    query: str,
    limit: int = 10,
) -> List[Document]

Document Metadata:

• source: "builtsimple-arxiv"
• arxiv_id: ArXiv identifier (e.g., "2301.12345")
• title: Paper title
• authors: Author list
• year: Publication year
• url: Link to ArXiv

BuiltSimpleWikipediaReader

BuiltSimpleWikipediaReader(
    api_key: Optional[str] = None,
    timeout: int = 30,
)

def load_data(
    query: str,
    limit: int = 10,
) -> List[Document]

Document Metadata:

• source: "builtsimple-wikipedia"
• title: Article title
• url: Link to Wikipedia

Rate Limits

Tier	Rate Limit	Notes
Free	10 req/min	No API key needed
Pro	100 req/min	Requires API key

Get an API key at pubmed.built-simple.ai or arxiv.built-simple.ai.

Why Built-Simple?

Unlike scraping or official APIs:

• Pre-indexed vectors - No embedding costs, instant semantic search
• Hybrid search - Combines BM25 + vector similarity
• Always available - No rate limit hell from upstream providers
• Structured data - Clean JSON responses with full metadata

Contributing

This package is part of the LlamaIndex ecosystem. To contribute:

1. Fork the repo
2. Create a feature branch
3. Submit a PR to run-llama/llama_index

License

MIT License - see LICENSE for details.

Links

• Built-Simple
• PubMed API Docs
• ArXiv API Docs
• LlamaIndex
• LlamaHub