lit-agent

Advanced academic identifier extraction and validation system with comprehensive assessment capabilities.

Features

• 🔍 Reference Extraction: Extract academic references from Deepsearch results in various formats
• 🔍 Multi-Phase Identifier Extraction: Extract DOI, PMID, and PMC identifiers from academic URLs using URL patterns, web scraping, and PDF text analysis
• 🎯 AI-Powered Topic Validation: LLM-based relevance assessment to ensure extracted papers match your research domain (e.g., astrocyte biology)
• 📊 Comprehensive Validation Pipeline: Multi-layered validation using format checking, NCBI API verification, and metapub integration
• 📈 Detailed Reporting & Visualization: Interactive HTML reports with charts, statistics, and actionable recommendations
• 🔬 Manual Review Guidance: Systematic sampling strategies and pause-point assessments for quality control
• 🤖 Unified LLM API: Support for OpenAI, Anthropic, and 100+ other providers via LiteLLM
• 📝 Multiple Citation Formats: Handle numbered citations ([1]), author-year (Smith et al., 2024), and plain URLs

Quick Start

Installation

# Clone the repository
git clone https://github.com/dosumis/lit_agent.git
cd lit_agent

# Install with uv (recommended)
uv sync --dev

# Or with pip
pip install -e ".[dev]"

API Keys Setup

# Copy example environment file
cp .env.example .env

# Edit .env and add your API keys
OPENAI_API_KEY=your_openai_key_here
ANTHROPIC_API_KEY=your_anthropic_key_here

# For academic identifier validation (required for validation features)
[email protected]  # Should be registered with NCBI
NCBI_API_KEY=your_ncbi_key        # Optional but recommended for higher rate limits

Usage

Bibliography → CSL-JSON mapping

Take a DeepSearch-style bibliography (URLs, optionally with source_id) and return CSL-JSON keyed by the original reference numbers:

from lit_agent.identifiers import resolve_bibliography

bibliography = [
    {"source_id": "1", "url": "https://pubmed.ncbi.nlm.nih.gov/37674083/"},
    {"source_id": "2", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11239014/"},
    {"source_id": "3", "url": "https://doi.org/10.1038/s41586-023-06502-w"},
]

result = resolve_bibliography(
    bibliography,
    validate=True,     # NCBI/metapub validation + metadata fetch
    scrape=False,      # Enable if you want web/PDF scraping for failures
    pdf=False,
    topic_validation=False,
)

print(result.citations["1"]["PMID"])     # "37674083"
print(result.citations["2"]["PMCID"])    # "PMC11239014"
print(result.citations["3"]["DOI"])      # "10.1038/s41586-023-06502-w"
print(result.citations["1"]["resolution"])  # methods, confidence, validation, errors

Each citation is CSL-JSON–compatible with a custom resolution block:

• id is the original source_id (or 1-based string if absent)
• URL, identifiers (DOI/PMID/PMCID), optional metadata (title, author, container-title, issued, etc.)
• resolution: confidence, methods, validation statuses, errors, source_url, optional canonical_id

Render to compact text with citeproc-py (optional dependency):

uv add --dev citeproc-py

from lit_agent.identifiers import render_bibliography_to_strings

rendered, meta = render_bibliography_to_strings(result, style="vancouver")
for line in rendered:
    print(line)  # e.g., "[1] Doe et al. 2024 Example Paper 10.1038/s41586-023-06502-w"

If citeproc-py is not installed, the helper falls back to a minimal compact formatter.

Academic Identifier Extraction

Extract DOI, PMID, and PMC identifiers from academic URLs with comprehensive validation:

from lit_agent.identifiers import extract_identifiers_from_bibliography

# Basic extraction from URLs
urls = [
    "https://pubmed.ncbi.nlm.nih.gov/37674083/",
    "https://www.nature.com/articles/s41586-023-06812-z",
    "https://pmc.ncbi.nlm.nih.gov/articles/PMC11239014/"
]

result = extract_identifiers_from_bibliography(
    urls=urls,
    use_web_scraping=True,      # Enable Phase 2 web scraping
    use_api_validation=True,    # Enable NCBI API validation
    use_topic_validation=True   # Enable LLM topic validation
)

print(f"Found {len(result.identifiers)} identifiers")
print(f"Success rate: {result.success_rate:.1%}")

Comprehensive Validation Assessment

Run a complete validation assessment with detailed reporting and visualizations:

from lit_agent.identifiers.validation_demo import run_validation_assessment_demo

# Run comprehensive validation assessment
report = run_validation_assessment_demo(
    urls=None,  # Uses default astrocyte biology test URLs
    use_topic_validation=True,
    output_dir="validation_reports",
    report_name="my_assessment"
)

# Check validation quality score
print(f"Validation Quality Score: {report['quality_score']}/100")

This generates:

• JSON Report: Complete validation statistics and metadata
• Text Summary: Human-readable assessment with recommendations
• CSV Export: Detailed paper information for spreadsheet analysis
• Interactive HTML: Visual dashboard with charts and insights
• Visualizations: 6 different chart types analyzing validation performance

Topic Validation for Specific Research Domains

Validate that extracted papers are relevant to your research topic:

from lit_agent.identifiers.topic_validator import TopicValidator

validator = TopicValidator()

# Validate a single identifier for astrocyte biology relevance
identifier = result.identifiers[0]
validation_result = validator.validate_identifier(identifier)

print(f"Relevant: {validation_result['is_relevant']}")
print(f"Confidence: {validation_result['confidence']}%")
print(f"Reasoning: {validation_result['reasoning']}")

Manual Review Workflow

The system provides systematic guidance for manual review:

# Generate paper classifications for manual review
from lit_agent.identifiers.reporting import ValidationReporter

reporter = ValidationReporter()
report = reporter.generate_validation_report(results, "manual_review")

# Papers needing manual review
classifications = report["paper_classifications"]
needs_review = classifications["needs_manual_review"]
low_confidence = classifications["low_confidence_relevant"]

print(f"Papers requiring manual review: {len(needs_review)}")
print(f"Low confidence papers: {len(low_confidence)}")

LLM Integration

Use the unified LLM API for custom analyses:

from lit_agent.agent_connection import create_agent_from_env

# Create agents from environment variables
agent = create_agent_from_env("anthropic")
response = agent.query("Analyze this paper abstract for astrocyte biology relevance...")

Command Line Interface

Run validation assessments directly from the command line:

# Run demo with default astrocyte biology URLs
uv run python -m lit_agent.identifiers.validation_demo

# Or run with Python directly
python src/lit_agent/identifiers/validation_demo.py

# Check the generated reports
ls demo_reports/
# validation_demo_20241105_143022.json
# validation_demo_20241105_143022_summary.txt
# validation_demo_20241105_143022_papers.csv
# validation_demo_20241105_143022.html

The demo script provides:

• ✅ Real-time Progress: Live updates on extraction and validation progress
• 📊 Immediate Results: Success rates, identifier counts, and confidence distributions
• 🎯 Topic Analysis: Relevance assessment for astrocyte biology research
• 💡 Actionable Recommendations: Specific suggestions for quality improvement
• 🌐 Interactive Reports: HTML dashboard with embedded visualizations

Troubleshooting

Common Issues

1. NCBI API Rate Limiting

# Error: HTTPSConnectionPool... Read timed out
# Solution: Add NCBI API key and email to .env
[email protected]
NCBI_API_KEY=your_ncbi_key

2. LLM API Errors

# Error: No API key provided
# Solution: Verify your .env file has the correct keys
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

3. Missing Dependencies

# Error: No module named 'matplotlib'
# Solution: Install visualization dependencies
uv sync --dev
# or
pip install matplotlib beautifulsoup4 pypdf lxml

4. Low Validation Quality Scores

• Check Topic Validation: Ensure your research domain matches the built-in astrocyte biology validation
• Review URLs: Verify input URLs are from academic sources
• API Connectivity: Confirm NCBI API access is working
• Manual Review: Use the paper classifications to identify systematic issues

Performance Optimization

For Large URL Lists:

• Enable caching for topic validation results
• Use batch processing for NCBI API calls
• Consider running validation in parallel chunks
• Monitor API rate limits and adjust delays

For Custom Research Domains:

• Modify the topic validation prompts in topic_validator.py
• Update keyword lists for your specific field
• Adjust confidence thresholds based on domain expertise

Development

Setup

# Install development dependencies
uv sync --dev

### Testing

```bash
# Run all tests (currently only tests required, code quality checks paused)
uv run pytest

# Run only unit tests (fast)
uv run pytest -m unit

# Run integration tests (requires API keys)
uv run pytest -m integration

# Run with coverage
uv run pytest --cov

Testing Strategy

This project follows strict Test-Driven Development with real integration testing:

• Unit Tests: Fast, isolated tests with mocks
• Integration Tests: Real API calls when keys available, mock fallback with warnings
• No Mocks for Integration: Real API testing is prioritized, mocks only as fallback
• Coverage tracking optional for now

Integration Test Behavior

# With API keys: Real API calls
--- Anthropic Hello World Response (REAL API) ---
The first recorded use of "Hello, World!" to demonstrate a programming language...

# Without API keys: Mock fallback with warning
UserWarning: ANTHROPIC_API_KEY not found - falling back to mock test.
--- Anthropic Hello World Response (MOCK) ---

Validation Assessment Features

Multi-Phase Extraction Pipeline

1. Phase 1 - URL Pattern Extraction: Fast extraction using regex patterns for known journal URLs
2. Phase 2 - Web Scraping: BeautifulSoup-based scraping for meta tags and JSON-LD data
3. Phase 3 - PDF Text Analysis: LLM-powered extraction from PDF content when available

Validation Methods

• Format Validation: Verify identifier formats (DOI, PMID, PMC patterns)
• NCBI API Validation: Real-time verification against PubMed database with metadata retrieval
• Metapub Integration: Cross-validation using metapub library
• Topic Validation: LLM-based assessment of paper relevance to research domains

Reporting & Analytics

• Comprehensive Statistics: Success rates, processing times, confidence distributions
• Interactive Visualizations: 6 chart types including confidence histograms, method comparisons, and topic analysis
• Quality Scoring: Data-driven assessment with actionable recommendations
• Manual Review Guidance: Stratified sampling strategies based on confidence scores

Pause-Point Assessment

The system provides systematic checkpoints for quality control:

• Validation Quality Score: 0-100 rating based on relevance, confidence, and success rates
• Automated Recommendations: Specific suggestions for improving extraction quality
• Paper Classifications: Systematic categorization for manual review prioritization
• Statistical Robustness: Confidence intervals and sample size recommendations

Architecture

• LiteLLM Integration: Unified API for 100+ LLM providers
• Environment-based Configuration: API keys via dotenv
• Modular Design: Abstract base classes with concrete implementations
• Error Handling: Comprehensive error handling with meaningful messages

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Write tests first (TDD approach)
4. Implement the feature
5. Ensure all tests pass (uv run pytest)
6. (Optional) Run additional checks later if re-enabled
7. Commit your changes (git commit -m 'Add amazing feature')
8. Push to the branch (git push origin feature/amazing-feature)
9. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Built with LiteLLM for unified LLM API access
• Uses uv for fast Python package management
• Code quality checks are currently paused (focus on tests)