Methodology & Agent Ecosystem

7-agent architecture • Real-time data integration • Automated research pipeline
🤖 7-Agent Architecture

Specialized AI agents working in concert to transform feedback into actionable intelligence:

C
Classifier
Input Categorization
Model:
DeepSeek V3
Role:
Processes transcripts, extracts facts, categorizes feedback
Output:
Tagged facts with confidence scores
R
Researcher
Intelligence Gathering
Model:
Gemini 2.5 Flash
Role:
Generates buyer pages, valuation comparables, market research
Output:
68+ buyer analyses, market intelligence
✓
Certifier
Quality Assurance
Model:
DeepSeek V3
Role:
Validates research against client thesis, checks alignment
Output:
Certified research packages
N
Nurturer
Relationship Management
Model:
GPT-4o Mini
Role:
Scheduled follow-ups, contact management, touchpoint planning
Output:
Relationship timelines, outreach schedules
V
Validator
Fact Checking
Model:
DeepSeek V3
Role:
Cross-checks financial data, verifies sources, audits claims
Output:
Verified intelligence packages
A
Auditor
System Integrity
Model:
DeepSeek V3
Role:
Audits entire pipeline, checks for gaps, ensures compliance
Output:
Audit reports, gap analysis
E
Executor
Output Delivery
Model:
GPT-4o Mini
Role:
Deploys outputs to deal room, updates dashboards, manages delivery
Output:
Live deal rooms, updated dashboards
🔄 Research Pipeline Flow

How feedback moves through our 7-agent system:

1
Transcript Ingestion
Fireflies.ai capture → Automated transcription → Sentiment analysis
2
Classifier Processing
Fact extraction → Categorization → Confidence scoring → Data model update
3
Research Activation
Automated task generation → Buyer research → Valuation analysis → Market intelligence
4
Quality Assurance
Certifier validation → Fact checking → Thesis alignment → Gap analysis
5
Output Delivery
Deal room deployment → Dashboard updates → Client notification → Audit trail
🔬 Overnight Research Process

$100 stress test methodology for HR.com overnight build:

Phase 1: Benchmark
Scrape Lovable.ai as quality benchmark • Build scoring rubric • Set evaluation criteria
Phase 2: Scoring
Score all 68 buyer pages • Rank by strategic fit • Categorize by buyer type
Phase 3: Comparison
Lovable vs CRM analysis • Gap identification • Prompt recommendations
Phase 4: Tournament
Bottom 12 buyers re-enriched • Agent performance ranking • Quality validation
Phase 5: Build
Overnight document generation • 212 outputs created • Full deal room deployment
Phase 6: Ranking
Agent performance evaluation • Cost efficiency analysis • Quality scoring
Phase 7: Inspection
Comprehensive audit report • Gap analysis • Improvement recommendations
Phase 8: Deployment
Git commit • Vercel deployment • Live site updates • Client notification
📊 Real-Time Data Integration
Supabase
201 targets • 68 buyers • Real-time sync
Fireflies.ai
Transcript processing • Sentiment analysis • Entity extraction
GitHub
Version control • Collaboration • Deployment pipeline
Vercel
Instant deployment • Global CDN • Auto-scaling
OpenRouter
7 AI models • Cost optimization • Performance tracking
Lob API
Physical letters • Tracking • Webhook integration
← Back to Debbie's Fact Dashboard View HR.com Deal Room