def extract_competitive_data(
    documents: List[Dict[str, Any]],
    use_llm: bool = True
) -> CompetitiveData:
    """Extract and synthesize competitive data from multiple documents."""

def extract_from_pdf(file_path: Path, use_llm: bool = True) -> Optional[Dict[str, Any]]:
    """Extract competitive data from PDF battlecard or analysis document."""

def synthesize_battlecards(extractions: List[Dict[str, Any]]) -> CompetitiveData:
    """Synthesize multiple per-competitor battlecards into unified competitive landscape."""

def _extract_battlecard_with_llm(filename: str, content: str) -> Dict[str, Any]:
    """Use LLM to extract structured data from battlecard."""

CompetitiveData = {
    "competitors": List[CompetitorEntry],  # Individual competitor profiles
    "key_differentiators": List[str],      # What sets company apart
    "winning_angles": List[str],           # Sales talking points
    "discovery_questions": List[str],      # Questions to ask prospects
    "market_positioning": str,             # Summary positioning statement
    "swot": SWOTAnalysis,                  # Company's SWOT analysis
    "source_documents": List[str],         # Source PDF filenames
}

CompetitorEntry = {
    "name": str,
    "description": str,
    "website": Optional[str],
    "strengths": List[str],
    "weaknesses": List[str],
    "threat_level": Literal["High", "Medium", "Low"],
    "feature_comparison": Dict[str, bool],
}

## Competitor Overview

| Competitor | Threat Level | Strengths | Weaknesses |
|------------|--------------|-----------|------------|
| SphereCo | 🟡 Medium | 3 | 10 |
| Silverfort | 🔴 High | 9 | 9 |
| Axonius | 🔴 High | 10 | 11 |
...

## Detailed Competitor Profiles

### Silverfort 🔴

Identity threat detection and response (ITDR) platform designed to catch
attackers at authentication runtime enforcement...

**Strengths:**
- Patented agentless inline enforcement—universal MFA and deny/step-up...
- Strong remediating options at runtime...
...

**Weaknesses:**
- Connectors for AD/Entra integrate with IAM/SIEM/PAM ecosystems, but not positioned as a general-purpose 'universal collector'...
...

def _run_extractors(inventory: list, use_llm: bool = True) -> dict:
    """Run specialized extractors on classified documents."""
    from .extractors import extract_competitive_data

    results = {...}

    # Group documents by type
    docs_by_type = group_by_document_type(inventory)

    # Run competitive extractor
    if "competitive_analysis" in docs_by_type:
        comp_docs = docs_by_type["competitive_analysis"]
        results["competitive"] = extract_competitive_data(comp_docs, use_llm=use_llm)

    return results

# Activate virtual environment
source .venv/bin/activate

# Run dataroom analyzer (LLM extraction enabled by default)
python -m src.agents.dataroom.analyzer "data/Secure-Inputs/Hydden Dataroom" "Hydden"

# Output:
# ============================================================
# DATAROOM ANALYZER
# ============================================================
# Company: Hydden
# Path: data/Secure-Inputs/Hydden Dataroom
# ============================================================
#
# 📁 Scanning dataroom...
#    Found 20 documents
#
# 🏷️  Classifying documents...
#    Classified 20 documents
#    High confidence: 20
#
# 🔍 Extracting competitive data from 9 documents...
#    ✓ Found 8 competitors
#    📄 Saved: 0-dataroom-inventory.json
#    📄 Saved: 0-dataroom-inventory.md
#    📄 Saved: 1-competitive-analysis.json
#    📄 Saved: 1-competitive-analysis.md
#
# ============================================================
# ✓ Analysis complete in 253.4s
# ✓ Artifacts saved to: output/Hydden-v0.0.1
# ============================================================