Case Study: Financial Analysis with Ensemble Verification¶
This case study covers designing a high-reliability AI system for generating equity research reports where accuracy is critical.
Table of Contents¶
- Problem Statement
- Requirements Analysis
- Architecture Design
- Ensemble Pipeline
- Fact Verification
- Quality Gates
- Results and Metrics
- Interview Walkthrough
Problem Statement¶
Company: Investment firm generating equity research reports
Challenge: - Reports influence multi-million dollar investment decisions - Zero tolerance for hallucinated financial data - Regulatory scrutiny on AI-generated analysis - Current manual process: 8 hours per report, $500 cost
Goal: - Reduce report generation time to < 30 minutes - Maintain accuracy at 99.5%+ - Clear audit trail for compliance - Cost target: < $50 per report
Requirements Analysis¶
Accuracy Requirements¶
| Data Type | Tolerance | Verification Method |
|---|---|---|
| Financial metrics (EPS, PE) | 0% error | Source verification |
| Percentage changes | ±0.1% | Cross-validation |
| Date references | 100% accuracy | Source extraction |
| Company names | 100% accuracy | Entity matching |
| Analyst quotes | Verbatim or flagged | Quote extraction |
Compliance Requirements¶
- All claims must cite source documents
- No forward-looking statements without disclaimers
- Clear AI-generated disclosure
- Full audit trail of generation process
- Human review for publication
Architecture Design¶
High-Level Pipeline¶
┌─────────────────────────────────────────────────────────────────┐
│ FINANCIAL ANALYSIS PIPELINE │
├─────────────────────────────────────────────────────────────────┤
│ │
│ Stage 1: Data Extraction (Self-Consistency k=5) │
│ └── Extract key metrics from filings with majority vote │
│ │
│ Stage 2: Analysis Generation (Mixture of Agents) │
│ ├── Model A: Quantitative analysis focus │
│ ├── Model B: Qualitative/narrative focus │
│ ├── Model C: Risk factor analysis │
│ └── Aggregator: Synthesize into coherent report │
│ │
│ Stage 3: Fact Verification (Multi-Agent Debate) │
│ └── 3 models debate each factual claim, flag disagreements │
│ │
│ Stage 4: Final Review (Panel of Judges) │
│ └── Quality score determines auto-publish vs human review │
│ │
└─────────────────────────────────────────────────────────────────┘
The pipeline as a flow. Each stage uses a different model class on purpose: extraction wants multimodal (charts and tables), generation wants narrative quality, audit wants reasoning depth, panel wants cheap-but-many for diversity:
flowchart LR
S1[Stage 1: Extraction<br/>Gemini 3 Pro<br/>Self-Consistency k=5] --> S2
S2[Stage 2: Analysis<br/>Mixture of Agents<br/>Quant + Narrative + Risk] --> S3
S3[Stage 3: Verification<br/>Multi-Agent Debate<br/>3 models per claim] --> S4
S4[Stage 4: Final Review<br/>Panel of Judges<br/>Quality score] --> D{Auto-publish<br/>threshold met}
D -->|yes| P[Publish]
D -->|no| H[Human Review Queue]Data Flow¶
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ 10-K/Q │ │ Earnings │ │ Analyst │
│ Filings │ │ Calls │ │ Reports │
└──────┬──────┘ └──────┬──────┘ └──────┬──────┘
│ │ │
└───────────────────┴───────────────────┘
│
▼
┌───────────────┐
│ Data │
│ Ingestion │
└───────┬───────┘
│
▼
┌───────────────┐
│ Extraction │
│ (k=5 SC) │
└───────┬───────┘
│
▼
┌────────────┴────────────┐
│ Structured Data │
│ (verified metrics) │
└────────────┬────────────┘
│
▼
┌───────────────┐
│ MoA │
│ Generation │
└───────┬───────┘
│
▼
┌───────────────┐
│ Debate │
│ Verification │
└───────┬───────┘
│
▼
┌───────────────┐
│ Panel │
│ Review │
└───────┬───────┘
│
┌───────────┴───────────┐
▼ ▼
┌─────────────┐ ┌─────────────┐
│ Auto-Publish│ │Human Review │
│ (high conf) │ │ (low conf) │
└─────────────┘ └─────────────┘
The data lineage in Mermaid, showing how three input sources converge into one verified output:
flowchart TD
F1[10-K and 10-Q Filings] --> ING[Data Ingestion]
F2[Earnings Calls] --> ING
F3[Analyst Reports] --> ING
ING --> EX[Extraction<br/>k=5 Self-Consistency]
EX --> SD[(Structured Data<br/>verified metrics)]
SD --> MOA[MoA Generation<br/>3 specialized agents]
MOA --> DEB[Debate Verification<br/>flag disagreements]
DEB --> PAN[Panel Review<br/>quality score]
PAN --> AP[Auto-Publish<br/>high confidence]
PAN --> HR[Human Review<br/>low confidence]Ensemble Pipeline¶
Stage 1: Multimodal Data Extraction (Gemini 3 Pro)¶
class FinancialDataExtractor:
"""
Using Gemini 3 Pro to handle complex 10-K tables and charts natively.
"""
async def extract_metrics(self, doc_pages: list[bytes]) -> dict:
# Gemini 3 Pro processes charts/tables as images + text natively
response = await genai.GenerativeModel("gemini-3.0-pro").generate_content(
[{"text": "Extract all balance sheet items into JSON."}, *doc_pages]
)
return json.loads(response.text)
Stage 2: Analysis Generation (Claude 4.5 Opus)¶
class AnalysisEngine:
"""
Claude 4.5 Opus for deep qualitative synthesis and narrative coherence.
"""
async def generate_report(self, data: dict) -> str:
# High-cost, high-reliability generation for equity research
return await self.anthropic.messages.create(
model="claude-4.5-opus-20251101",
messages=[{"role": "user", "content": f"Analyze: {data}"}]
)
Stage 3: Audit & Verification (o3 Reasoning Model)¶
class AuditorAgent:
"""
Using o3 (OpenAI) with high reasoning budget to audit claims.
Thinking mode is used to detect subtle accounting contradictions.
"""
async def audit_claim(self, claim: str, raw_data: str) -> dict:
# o3 'Thinking' mode enables deep logical inference over financial data
response = await self.openai.chat.completions.create(
model="o3-2025-12",
reasoning_effort="high",
messages=[{"role": "user", "content": f"Find any contradiction in: {claim} vs {raw_data}"}]
)
return self.parse_audit(response)
Stage 3: Fact Verification with Multi-Agent Debate¶
The debate stage is what catches the subtle hallucinations a single model misses. Three independent debaters verify each claim in parallel; consensus wins, dissent flags the claim for human review:
sequenceDiagram
participant CE as Claim Extractor
participant D1 as Debater A<br/>Claude 4.5 Opus
participant D2 as Debater B<br/>GPT-5.2
participant D3 as Debater C<br/>Gemini 3 Pro
participant CON as Consensus Logic
participant OUT as Verification Result
CE->>CE: extract factual claims<br/>from report
Note over CE,D3: For each claim, debaters verify independently
par Independent verification
CE->>D1: claim + source docs
D1-->>CON: verdict (supported/inferred/unsupported/contradicted)
and
CE->>D2: claim + source docs
D2-->>CON: verdict
and
CE->>D3: claim + source docs
D3-->>CON: verdict
end
CON->>CON: check consensus
alt all agree supported
CON->>OUT: verified
else any contradiction
CON->>OUT: flagged for human review
else split verdicts
CON->>OUT: low confidence
endclass FactVerificationDebate:
"""
Extract claims from the report and have multiple models
debate their accuracy.
"""
def __init__(self, debaters: list, rounds: int = 2):
self.debaters = debaters
self.rounds = rounds
self.claim_extractor = ClaimExtractor()
async def verify_report(self, report: str, source_docs: list[str]) -> dict:
# Extract factual claims
claims = await self.claim_extractor.extract(report)
verification_results = []
for claim in claims:
result = await self.debate_claim(claim, source_docs)
verification_results.append(result)
return {
"verified_claims": [r for r in verification_results if r["verified"]],
"disputed_claims": [r for r in verification_results if not r["verified"]],
"overall_confidence": self.calculate_confidence(verification_results)
}
async def debate_claim(self, claim: dict, source_docs: list[str]) -> dict:
verification_prompt = f"""
Verify this claim against the source documents.
Claim: {claim['text']}
Source documents:
{self.format_sources(source_docs)}
Is this claim:
1. Supported: Explicitly stated in sources
2. Inferred: Reasonably derived from sources
3. Unsupported: Not found in sources
4. Contradicted: Conflicts with sources
Provide your verdict with evidence.
"""
# Each debater verifies independently
verdicts = await asyncio.gather(*[
debater.generate(verification_prompt)
for debater in self.debaters
])
# Check consensus
parsed_verdicts = [self.parse_verdict(v) for v in verdicts]
consensus = self.check_consensus(parsed_verdicts)
return {
"claim": claim,
"verified": consensus["agreed"] and consensus["verdict"] in ["supported", "inferred"],
"confidence": consensus["agreement_ratio"],
"verdicts": parsed_verdicts
}
Quality Gates¶
Automated Quality Checks¶
class QualityGate:
def __init__(self):
self.thresholds = {
"claim_verification_rate": 0.95, # 95% claims verified
"data_accuracy": 0.99, # 99% metrics accurate
"panel_score": 4.0, # 4/5 minimum
"disputed_claims_max": 2 # Max 2 disputed claims
}
async def evaluate(self, report_data: dict) -> dict:
checks = {}
# Check claim verification rate
verified_rate = len(report_data["verified_claims"]) / len(report_data["all_claims"])
checks["claim_verification"] = {
"passed": verified_rate >= self.thresholds["claim_verification_rate"],
"value": verified_rate,
"threshold": self.thresholds["claim_verification_rate"]
}
# Check data accuracy
data_accuracy = report_data["extraction_accuracy"]
checks["data_accuracy"] = {
"passed": data_accuracy >= self.thresholds["data_accuracy"],
"value": data_accuracy,
"threshold": self.thresholds["data_accuracy"]
}
# Check panel score
panel_score = report_data["panel_score"]
checks["panel_score"] = {
"passed": panel_score >= self.thresholds["panel_score"],
"value": panel_score,
"threshold": self.thresholds["panel_score"]
}
# Determine routing
all_passed = all(c["passed"] for c in checks.values())
return {
"checks": checks,
"routing": "auto_publish" if all_passed else "human_review",
"disputed_claims": report_data["disputed_claims"]
}
Human Review Interface¶
class HumanReviewQueue:
async def queue_for_review(self, report: dict, quality_result: dict):
review_item = {
"report_id": report["id"],
"report_content": report["content"],
"disputed_claims": quality_result["disputed_claims"],
"quality_checks": quality_result["checks"],
"sources": report["sources"],
"priority": self.calculate_priority(quality_result),
"queued_at": datetime.now()
}
await self.review_queue.enqueue(review_item)
# Notify reviewers
await self.notify_reviewers(review_item)
Results and Metrics¶
Performance Comparison¶
| Metric | Manual Process | AI Pipeline | Improvement |
|---|---|---|---|
| Time per report | 8 hours | 25 minutes | 19x faster |
| Cost per report | $500 | $42 | 92% reduction |
| Factual error rate | 2.1% | 0.4% | 81% reduction |
| Human review load | 100% | 28% | 72% reduction |
Quality Metrics¶
| Quality Dimension | Target | Achieved |
|---|---|---|
| Data extraction accuracy | 99% | 99.3% |
| Claim verification rate | 95% | 96.8% |
| Panel quality score | 4.0/5.0 | 4.2/5.0 |
| Regulatory compliance | 100% | 100% |
Cost Breakdown (Dec 2025)¶
| Component | Cost | Percentage |
|---|---|---|
| Data extraction (Gemini 3 Pro) | $5 | 11% |
| Analysis (Claude 4.5 Opus) | $20 | 44% |
| o3 Thinking-Audit (High) | $15 | 33% |
| Infrastructure & Vector Ops | $5 | 12% |
| Total | $45 | 100% |
Note: o3 auditing represents 33% of the cost but catches 98% of hallucinations that Claude 4.5 misses, justifying the 'Thinking' token premium.
Interview Walkthrough¶
Interviewer: "Design an AI system for generating financial research reports with very high accuracy requirements."
Strong response:
- Clarify accuracy requirements (1 min)
- "What's the acceptable error rate for financial data?"
- "What's the regulatory compliance requirement?"
-
"Is latency or accuracy the priority?"
-
Acknowledge the core challenge (1 min)
-
"The key challenge is that hallucinations are unacceptable for financial data. A single wrong number could mislead investment decisions. I need ensemble methods for reliability."
-
High-level architecture (3 min)
- "I would use a multi-stage pipeline with different ensemble techniques at each stage:"
- "Data extraction: Self-consistency with k=5 for unanimous agreement on numbers"
- "Analysis: Mixture of Agents for diverse perspectives"
- "Verification: Multi-agent debate to catch hallucinations"
-
"Quality gate: Panel of judges to score before publishing"
-
Deep dive on fact verification (3 min)
- "For fact verification, I extract every factual claim from the report"
- "Three diverse models debate whether each claim is supported by sources"
- "If they disagree, the claim is flagged for human review"
-
"This catches subtle errors that single-model verification misses"
-
Cost-quality tradeoff (2 min)
- "This pipeline is 10-20x more expensive than single-model generation"
- "But for financial reports, the cost of errors (legal, reputational) far exceeds the cost of verification"
-
"I would implement confidence-based routing: auto-publish high-confidence reports, human-review low-confidence ones"
-
Monitoring (1 min)
- "I would track extraction accuracy, claim verification rate, and panel scores continuously"
- "Drift detection would alert if accuracy drops"
- "Full audit trail for compliance"
Key Learnings¶
-
Self-consistency alone is insufficient for numerical data extraction. Unanimous agreement (k/k votes) should be required.
-
Multi-agent debate most effective for catching subtle reasoning errors and hallucinations.
-
Source attribution is critical for both accuracy and compliance. Every claim must link to source documents.
-
Confidence-based routing is essential for cost management. Not every report needs full ensemble verification.
-
Human-in-the-loop is still necessary for disputed claims and edge cases. Design for graceful escalation.
References¶
- Verga et al. "Replacing Judges with Juries: Evaluating LLM Generations with a Panel of Diverse Models" (2024)
- Du et al. "Improving Factuality and Reasoning in Language Models through Multiagent Debate" (2023)
- SEC AI Disclosure Requirements: https://www.sec.gov/