semantic-kernel-builtsimple

Semantic Kernel plugins for Built-Simple research APIs, providing easy access to PubMed, ArXiv, and Wikipedia for AI-powered research assistants.

Features

• PubMed Plugin - Search 4.5M+ peer-reviewed biomedical articles with full text support
• ArXiv Plugin - Search 2.7M+ preprints in physics, math, CS, ML, and AI
• Wikipedia Plugin - Semantic search over Wikipedia for general knowledge
• Combined Research Plugin - All sources in one plugin for comprehensive research
• Async-first - Built with async/await for optimal performance
• Function calling ready - Works with OpenAI, Azure OpenAI, and other LLM providers

What Data is Included

PubMed Results

• Title, abstract, and full article text when available
• Journal name and publication year
• PMID and DOI identifiers
• Direct links to PubMed and DOI URLs
• Author information

ArXiv Results

• Paper title and abstract
• Author list with affiliations
• ArXiv ID with links to abstract and PDF
• Publication year and categories (cs.AI, physics, math, etc.)

Wikipedia Results

• Article title and content summary
• Direct Wikipedia URLs
• Category information

Installation

pip install semantic-kernel-builtsimple

Quick Start

Basic Plugin Usage

import asyncio
from semantic_kernel import Kernel
from semantic_kernel_builtsimple import (
    BuiltSimplePubMedPlugin,
    BuiltSimpleArxivPlugin,
    BuiltSimpleWikipediaPlugin,
)

async def main():
    kernel = Kernel()
    
    # Add plugins
    kernel.add_plugin(BuiltSimplePubMedPlugin(), plugin_name="pubmed")
    kernel.add_plugin(BuiltSimpleArxivPlugin(), plugin_name="arxiv")
    kernel.add_plugin(BuiltSimpleWikipediaPlugin(), plugin_name="wikipedia")
    
    # Invoke a function directly
    result = await kernel.invoke(
        plugin_name="pubmed",
        function_name="search_pubmed",
        query="CRISPR gene therapy clinical trials",
        limit=3,
    )
    print(result)

asyncio.run(main())

With OpenAI Function Calling

import asyncio
from semantic_kernel import Kernel
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion
from semantic_kernel.connectors.ai.function_choice_behavior import FunctionChoiceBehavior
from semantic_kernel.contents.chat_history import ChatHistory
from semantic_kernel_builtsimple import BuiltSimpleResearchPlugin

async def main():
    kernel = Kernel()
    
    # Add AI service
    kernel.add_service(OpenAIChatCompletion(
        service_id="chat",
        ai_model_id="gpt-4o",
    ))
    
    # Add research plugin (includes PubMed, ArXiv, Wikipedia)
    kernel.add_plugin(BuiltSimpleResearchPlugin(), plugin_name="research")
    
    # Get execution settings with function calling enabled
    settings = kernel.get_prompt_execution_settings_from_service_id("chat")
    settings.function_choice_behavior = FunctionChoiceBehavior.Auto(
        filters={"included_plugins": ["research"]}
    )
    
    # Ask a research question - the AI will automatically use the plugins
    result = await kernel.invoke_prompt(
        prompt="What are the latest advances in transformer architectures for natural language processing? Search both ArXiv for recent papers and PubMed for any clinical applications.",
        settings=settings,
    )
    print(result)

asyncio.run(main())

Using the Combined Research Plugin

import asyncio
from semantic_kernel import Kernel
from semantic_kernel_builtsimple import BuiltSimpleResearchPlugin

async def main():
    kernel = Kernel()
    
    # Single plugin provides access to all sources
    kernel.add_plugin(BuiltSimpleResearchPlugin(), plugin_name="research")
    
    # Search all sources at once
    result = await kernel.invoke(
        plugin_name="research",
        function_name="search_all_sources",
        query="artificial intelligence in drug discovery",
        limit_per_source=3,
    )
    print(result)

asyncio.run(main())

With Azure OpenAI

import asyncio
from semantic_kernel import Kernel
from semantic_kernel.connectors.ai.open_ai import AzureChatCompletion
from semantic_kernel.connectors.ai.function_choice_behavior import FunctionChoiceBehavior
from semantic_kernel_builtsimple import BuiltSimplePubMedPlugin, BuiltSimpleArxivPlugin

async def main():
    kernel = Kernel()
    
    # Add Azure OpenAI service
    kernel.add_service(AzureChatCompletion(
        service_id="azure_chat",
        deployment_name="gpt-4o",
        endpoint="https://your-resource.openai.azure.com/",
        api_key="your-api-key",
    ))
    
    # Add research plugins
    kernel.add_plugin(BuiltSimplePubMedPlugin(), plugin_name="pubmed")
    kernel.add_plugin(BuiltSimpleArxivPlugin(), plugin_name="arxiv")
    
    # Enable function calling
    settings = kernel.get_prompt_execution_settings_from_service_id("azure_chat")
    settings.function_choice_behavior = FunctionChoiceBehavior.Auto()
    
    result = await kernel.invoke_prompt(
        prompt="Find recent research on mRNA vaccines for cancer treatment",
        settings=settings,
    )
    print(result)

asyncio.run(main())

Plugin Reference

BuiltSimplePubMedPlugin

Searches PubMed biomedical literature database.

Functions:

• search_pubmed(query, limit=5) - Search for papers
• get_pubmed_full_text(pmid) - Get full article text by PMID

plugin = BuiltSimplePubMedPlugin(
    base_url="https://pubmed.built-simple.ai",  # optional override
    api_key=None,  # optional for higher rate limits
    timeout=30.0,  # request timeout in seconds
)

BuiltSimpleArxivPlugin

Searches ArXiv preprint server.

Functions:

• search_arxiv(query, limit=5) - Search for preprints

plugin = BuiltSimpleArxivPlugin(
    base_url="https://arxiv.built-simple.ai",
    api_key=None,
    timeout=30.0,
)

BuiltSimpleWikipediaPlugin

Searches Wikipedia articles semantically.

Functions:

• search_wikipedia(query, limit=5) - Search for articles

plugin = BuiltSimpleWikipediaPlugin(
    base_url="https://wikipedia.built-simple.ai",
    api_key=None,
    timeout=30.0,
)

BuiltSimpleResearchPlugin

Combined plugin with access to all sources.

Functions:

• search_pubmed(query, limit=5) - Search PubMed
• search_arxiv(query, limit=5) - Search ArXiv
• search_wikipedia(query, limit=5) - Search Wikipedia
• search_all_sources(query, limit_per_source=3) - Search all simultaneously

plugin = BuiltSimpleResearchPlugin(
    pubmed_url=None,  # optional overrides
    arxiv_url=None,
    wikipedia_url=None,
    api_key=None,
    timeout=30.0,
)

When to Use Each Source

Source	Best For
PubMed	Medical research, clinical studies, drug development, genomics, biology, healthcare
ArXiv	AI/ML papers, physics, mathematics, computer science, cutting-edge preprints
Wikipedia	General knowledge, definitions, historical facts, biographies, concepts

Example Use Cases

Research Assistant

Build an AI research assistant that can answer questions using scientific literature:

system_prompt = """You are a research assistant with access to:
- PubMed for biomedical literature
- ArXiv for physics/math/CS preprints  
- Wikipedia for general knowledge

When answering questions:
1. Search relevant sources based on the topic
2. Synthesize information from multiple papers
3. Always cite your sources with titles and IDs
"""

Literature Review

Help researchers quickly survey a topic:

result = await kernel.invoke(
    plugin_name="research",
    function_name="search_all_sources",
    query="attention mechanisms in neural networks",
    limit_per_source=10,
)

Fact-Checking

Verify claims with authoritative sources:

# Check Wikipedia for general facts
wiki = await kernel.invoke(
    plugin_name="wikipedia",
    function_name="search_wikipedia",
    query="discovery of penicillin",
)

# Verify with primary sources
pubmed = await kernel.invoke(
    plugin_name="pubmed", 
    function_name="search_pubmed",
    query="penicillin discovery Alexander Fleming",
)

Error Handling

The plugins handle errors gracefully and return descriptive messages:

# If API is unavailable or query fails
result = await kernel.invoke(
    plugin_name="pubmed",
    function_name="search_pubmed",
    query="test query",
)
# Returns: "Error searching PubMed: <error details>" instead of raising

For programmatic error handling, import the exception:

from semantic_kernel_builtsimple import BuiltSimpleAPIError

try:
    # Direct client usage
    async with BuiltSimpleClient("https://pubmed.built-simple.ai") as client:
        result = await client.post("/hybrid-search", data={"query": "test"})
except BuiltSimpleAPIError as e:
    print(f"API error: {e.message}, status: {e.status_code}")

License

MIT License - see LICENSE for details.

Links

• Built-Simple
• PubMed API
• ArXiv API
• Wikipedia API
• Semantic Kernel
• Semantic Kernel Documentation