Build a Pre-IPO Shareholding Audit Agent with Claude Agents SDK
Industrial tech major Zetwerk's draft red herring prospectus (UDRHP) disclosure highlighted complex shareholding structures in pre-IPO Indian tech companies. In this challenge, you will implement an automated prospectus audit agent using Claude Agents SDK. The agent ingests regulatory financial filings (DRHP, UDRHP), extracts cap table ownership structures, identifies ESOP pool allocations, flags related-party transactions, and verifies compliance with SEBI issue guidelines.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a financial prospectus audit agent using Claude Agents SDK to parse regulatory shareholding disclosures.
How work is evaluated
Evaluates cap table share percentage calculations and anomaly detection accuracy.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
cap_table_sum_test
Verifies promoter percentage is calculated accurately (4.0M / 10.0M = 40%).
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
extraction_precision
Accuracy in extracting share values from regulatory tables. • target: 0.98 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Utilize Claude Agents SDK extended thinking capabilities for complex document parsing.
Extract structured capitalization tables from unstructured financial filings.
Verify share dilution math and ESOP allocation boundaries.
Generate automated compliance reports highlighting SEBI disclosure red flags.
Reference links and supporting material
Annotated DRHP and UDRHP financial filings with ground truth cap tables and ESOP allocation tables.
Challenge setup
Evaluates cap table share percentage calculations and anomaly detection accuracy.
Challenge input
JSON array of shareholder entries with share_count and class_type.
Claude Agents SDK
Required framework for deep reasoning document extraction.
Evaluated output
JSON with total_shares, promoter_holding_pct, compliance_status.
- Verifies promoter percentage is calculated accurately (4.0M / 10.0M = 40%).
- Accuracy in extracting share values from regulatory tables. • target: 0.98 • range: 0-1
- Benchmark: Financial Document Extraction Benchmark
- Extraction Precision target: 0.98
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Agency (Observability) · Required
- Ax (Adaptive Experimentation) · Optional
- Cartesia · Optional
Evaluation contract
- cap_table_sum_test · Weight 1
- extraction_precision · Weight 1
Metrics
- extraction_precision · Target 0.98
Benchmarks
- Financial Document Extraction Benchmark
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON array of shareholder entries with share_count and class_type.
Output: JSON with total_shares, promoter_holding_pct, compliance_status.
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.