Research Automation¶

Archivus research automation conducts autonomous web research with real-time accuracy verification. Every finding is validated against both your trusted documents and live external sources for maximum reliability.

Dual Validation Architecture¶

Pro+

What Makes It Unique:

Most AI research tools either hallucinate or provide unverified web results. Archivus is the only platform with dual validation:

1. Internal Validation: Ground findings against your Source of Truth documents
2. External Validation: Verify via Perplexity Sonar against live web data

Example Validation Flow:

Research Finding: "Company X acquired Company Y for $2B"

↓ Internal Validation (Phase 4)
├─ Compared against your trusted documents
├─ Grounding Score: 87%
└─ Notes: "Aligns with internal M&A briefing"

↓ External Validation (Phase 4B)
├─ Cross-referenced against live web
├─ Status: "verified" (95% confidence)
├─ Correction: "Final value was $2.1B"
└─ Citations: [reuters.com, techcrunch.com, sec.gov]

Research Workflow¶

5-Phase Process:

Phase 1: Topic Analysis¶

• Parse research topic
• Generate targeted search queries
• Identify key entities and concepts
• Estimate required depth

Phase 2: Web Search¶

Search providers:

• Tavily: AI-optimized search API
• Serper: Google search API
• Perplexity: AI-powered search

Cost: 2 credits per query

Phase 3: Synthesis¶

• Analyze search results
• Extract key findings
• Identify claims and evidence
• Structure information

Phase 4: Internal Validation¶

Pro+

Ground findings against Source of Truth:

• Compare to trusted document collections
• Calculate grounding score (0-100%)
• Flag contradictions with internal docs
• Note supporting evidence

Cost: 1 credit per finding

Phase 5: External Validation¶

Pro+

Verify via Perplexity Sonar:

• Cross-reference against live web data
• Get validation status with confidence
• Receive corrections if outdated
• Obtain authoritative citations

Cost: 1 credit per finding

Validation Statuses¶

Each finding receives a validation status:

Research Reports¶

Two Report Modes:

Standard Mode¶

• Focused findings (5-10 key points)
• Internal validation only
• Quick turnaround (~2 minutes)
• Cost: ~10-15 credits

Power Mode¶

• Comprehensive findings (15-30 points)
• Dual validation (internal + external)
• Deep analysis (~5-10 minutes)
• Cost: ~25-40 credits

Report Format:

# Research Report: [Topic]
Generated: [Timestamp]
Sources Analyzed: [Count]
Validation Mode: [Internal + External]

## Executive Summary
[2-3 paragraph overview]

## Key Findings

### Finding 1: [Title]
**Status**: Verified ✓
**Confidence**: 95%

[Detailed description]

**Internal Validation**:
- Grounding Score: 89%
- Supporting Docs: M&A-Briefing-Q4.pdf

**External Validation**:
- Status: Verified
- Citations:
  - [reuters.com/article...]
  - [sec.gov/filing...]

### Finding 2: [Title]
**Status**: Contradicted ⚠️
**Confidence**: 87%

[Description with contradiction details]

**Internal Validation**:
- Conflicts with: Company-Policy-2025.pdf
- Recommended Action: Update policy

**External Validation**:
- Status: Verified (external source correct)
- Correction: [Updated information]

Source of Truth Integration¶

Pro+

Define trusted document collections for grounding:

Setup:

1. Create collection of authoritative documents
2. Mark collection as "Source of Truth"
3. Research automatically grounds against these docs
4. Contradictions flagged for review

Example Collections:

• Company policies and procedures
• Approved methodologies
• Compliance requirements
• Industry standards
• Historical precedents

Grounding Score:

Score = (Aligned Statements / Total Statements) × 100

95-100%: Fully aligned with Source of Truth
80-94%:  Mostly aligned, minor gaps
60-79%:  Partially aligned, review recommended
<60%:    Significant contradictions, caution advised

Research Use Cases¶

Competitive Intelligence¶

Topic: "AI document management competitors 2026"

Findings:
✓ Market size: $4.2B (verified)
✓ Top competitors identified (8 companies)
⚠️ Pricing data partially outdated
✓ Feature comparison current (validated)

Regulatory Research¶

Topic: "HIPAA compliance requirements for cloud storage"

Findings:
✓ Current regulations (verified via HHS.gov)
✓ Recent enforcement actions (verified)
⚠️ Internal policy gaps identified
✓ Best practices (validated externally)

Due Diligence¶

Topic: "Acme Corporation financial stability"

Findings:
✓ Revenue: $450M (verified via SEC filings)
⚠️ Internal projection differs (flagged)
✓ Recent acquisition (validated)
✓ Credit rating: A- (verified)

Market Research¶

Topic: "Remote work trends 2026"

Findings:
✓ 67% hybrid adoption (verified)
✓ Industry breakdown (validated)
⚠️ Contradicts internal survey (noted)
✓ Future projections (multiple sources)

API Access¶

Pro+

Programmatic research via API:

# Start research
POST /api/v1/research/start
{
  "topic": "AI regulations financial services",
  "mode": "power",
  "source_of_truth_ids": ["coll-uuid-1", "coll-uuid-2"],
  "external_validation": true
}

# Get research status
GET /api/v1/research/{research_id}/status

# Download report
GET /api/v1/research/{research_id}/report.pdf

Research Templates¶

Pre-configured research templates:

1. Competitive Analysis: Market sizing, competitor features, pricing
2. Regulatory Update: New regulations, enforcement, compliance gaps
3. Due Diligence: Company background, financials, risk factors
4. Market Research: Trends, statistics, projections
5. Technology Assessment: Features, adoption, vendors
6. Customer Research: Reviews, sentiment, pain points

Cost Management¶

Credit Usage Breakdown:

• Topic analysis: 3 credits
• Web search: 2 credits per query (typically 5-10 queries)
• Internal grounding: 1 credit per finding
• External validation: 1 credit per finding
• Report generation: 5 credits

Typical Research Costs:

• Simple topic (5 findings): 15-20 credits
• Standard research (10 findings): 25-35 credits
• Power research (20 findings): 45-65 credits

Optimization Tips:

• Use specific topics (reduces query count)
• Leverage Source of Truth (improves accuracy)
• Standard mode for quick research
• Power mode for critical decisions

Security & Compliance¶

Data Privacy:

• Research queries isolated by tenant
• Source of Truth access controls enforced
• External validation via secure APIs
• No sharing of internal documents

Audit Trail:

• Complete research history
• Query and result logging
• Validation status tracking
• Cost attribution per research

Getting Started¶

Web Interface:

1. Navigate to Research tab
2. Enter research topic
3. Select mode (Standard or Power)
4. Choose Source of Truth collections (optional)
5. Enable external validation (Pro+)
6. Start research
7. Review findings and download report

Best Practices:

• Be specific in research topics
• Define Source of Truth for your domain
• Review contradictions carefully
• Update internal docs when external validation corrects them
• Use Power mode for critical business decisions

View Research API Docs → See Research Examples →