Changelog - 2025-11-26 (#5)

Team Extractor - Phase 2 Complete!

Overview

Added the Team Extractor, completing all Phase 2 extractors (5/5). The dataroom analyzer now extracts comprehensive structured data from investment datarooms.

New Feature: Team Extractor

Purpose

Extract founder profiles, leadership team, headcount, advisors, and board members from pitch decks and team documents.

Implementation

Created src/agents/dataroom/extractors/team_extractor.py (~550 lines):

def extract_team_data(
    documents: List[Dict[str, Any]],
    use_llm: bool = True
) -> Optional[TeamData]:
    """Extract team data from classified documents."""

def extract_from_pdf(file_path: Path, doc_type: str, use_llm: bool = True) -> Optional[Dict[str, Any]]:
    """Extract team data from a PDF document."""

def extract_from_spreadsheet(file_path: Path, use_llm: bool = True) -> Optional[Dict[str, Any]]:
    """Extract team data from a spreadsheet (CSV/Excel)."""

def extract_from_text(file_path: Path, use_llm: bool = True) -> Optional[Dict[str, Any]]:
    """Extract team data from a text file."""

Extraction Approach

Three-stage extraction:

1. Rule-based parsing - Regex patterns for name/title combinations, headcount, LinkedIn URLs

2. Table extraction - pdfplumber for org charts and team tables

3. LLM enhancement - Claude Sonnet for detailed background extraction

Data extracted:

• Founders (name, title, LinkedIn, previous companies, roles, education, achievements, expertise, years experience)

• Leadership team (same structure as founders)

• Total headcount

• Headcount by department

• Advisors

• Board members

Testing Results (Hydden Dataroom)

Phase 2 Complete - All Extractors

The dataroom analyzer now outputs 6 numbered artifact files:

output/Company-v0.0.1/
├── 0-dataroom-inventory.json/md    # Document inventory & classification
├── 1-competitive-analysis.json/md  # Competitive landscape (battlecards)
├── 2-cap-table.json/md             # Ownership structure
├── 3-financial-analysis.json/md    # Financial projections
├── 4-traction-analysis.json/md     # Customers & traction metrics
└── 5-team-analysis.json/md         # Team & leadership profiles

Extractor Summary

New Report Formatter

Added format_team_report() to analyzer.py:

• Founder profiles with backgrounds, education, achievements

• Leadership team table

• Organization headcount by department

• Advisors and board members

• Extraction notes

Integration Updates

_run_extractors() Updated

Now calls team extractor for document types:

• team_bios

• pitch_deck

_extract_key_facts() Updated

Now extracts team facts:

• Founder count and names

• Notable prior companies

• Leadership count

• Total headcount

• Advisor count

• Board size

Bug Fix

Fixed format_team_report() crash when headcount_by_department contains None values - now filters out invalid entries before sorting.

Files Changed Summary

New Files (1)

Modified Files (2)

Phase 2 Complete

All Extractors Done

• Competitive Extractor - Battlecard synthesis

• Cap Table Extractor - Ownership structure

• Financial Extractor - Projections & metrics

• Traction Extractor - Customers & revenue

• Team Extractor - Founders & leadership

Next: Phase 3 - Data Synthesis & Memo Integration

• Cross-reference data between extractors

• Identify conflicts and data gaps

• Generate memo section drafts from extracted data

• Integrate with main investment memo orchestrator

Full Hydden Test Results

Complete analysis in 282.6 seconds:

Usage

# Activate virtual environment
source .venv/bin/activate

# Run dataroom analyzer
python -m src.agents.dataroom.analyzer "data/Secure-Inputs/Hydden Dataroom" "Hydden"

# Output:
# 📁 Scanning dataroom...
#    Found 20 documents
# 🏷️  Classifying documents...
#    Classified 20 documents
# 🔍 Extracting competitive data from 9 documents...
#    ✓ Found 8 competitors
# 📊 Extracting cap table data from 1 documents...
#    ✓ Found 5 shareholders
# 💰 Extracting financial data from 1 documents...
#    ✓ Found financial projections
# 📈 Extracting traction data from 2 documents...
#    ✓ Found: $3,000,000 ARR, $2,026 pipeline
# 👥 Extracting team data from 2 documents...
#    ✓ Found: 2 founders, 4 leaders, 17 total headcount
# ✓ Analysis complete in 282.6s

Contributors

Implementation by Claude Code (Anthropic) with user guidance.