Skip to main content
Back to challenges
Challenge

Agent for Auditable Financial Model Generation

To make financial modeling predictable and auditable, this challenge focuses on building an AI agent system using LlamaIndex for advanced financial analysis. Unlike traditional multi-agent tool-calling applications, this challenge emphasizes LlamaIndex's agentic capabilities for structured data processing, tool use, and complex reasoning without relying on tool orchestrations. Participants will design an agent that can ingest raw financial data (e.g., CSV, JSON), apply business logic, generate financial models, and produce comprehensive audit trails. The system will use GPT-5 for core reasoning and model generation, with Claude Sonnet 5 for summarization and clarification. Ray Serve and Novita AI will be leveraged for efficient and scalable deployment of these models, ensuring reliable inference. The agent will interact with simulated financial APIs and spreadsheet tools, producing auditable outputs that explain its reasoning and data transformations, enhancing trust and transparency in AI-driven financial insights.

Workflow AutomationHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Develop a LlamaIndex agent using GPT-5 and Claude Sonnet 5 for auditable financial modeling, leveraging tool use and model serving platforms.

Delivery guide

How work is evaluated

Evaluation

Evaluation will assess the LlamaIndex agent's accuracy in financial modeling, the completeness and clarity of its audit trails, and its ability to handle structured financial data.

Datasets

Challenge datasets

Public datasets, plus the private datasets you have access to.

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

Dimensions
3 scoring checks
Binary
3 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

Correct Model Calculation

The generated financial projections must be numerically accurate based on input data and parameters.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

Audit Trail Presence

The 'audit_trail' list must contain at least 3 distinct steps.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 3

Audit Trail Detail

The average length of individual entries in the 'audit_trail' (in characters). • target: 50 • range: 20-100

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

What you should walk away with

  • Master LlamaIndex's agent framework for building complex, tool-calling agents that interact with structured data sources (e.g., CSV, JSON).

  • Design and implement custom tools for LlamaIndex agents to perform financial calculations, interact with simulated spreadsheet APIs, and retrieve market data.

  • Integrate GPT-5 as the primary reasoning engine for the LlamaIndex agent, enabling complex financial planning, scenario analysis, and model generation.

  • Utilize Claude Sonnet 4 within the LlamaIndex agent for generating clear, concise summaries of financial reports and explaining modeling assumptions or audit trails.

  • Deploy LlamaIndex agents and their underlying GPT-5 and Claude Sonnet 4 models using Ray Serve for scalable and fault-tolerant inference serving.

  • Leverage Novita AI for optimizing the deployment and runtime performance of the AI models, ensuring high throughput for financial analysis tasks.

  • Develop strategies for generating comprehensive audit trails within the LlamaIndex agent's workflow, detailing data transformations, model choices, and decision-making steps.

  • Implement validation checks within the LlamaIndex agent to ensure the predictability and accuracy of generated financial models.

Resources and assets

Reference links and supporting material

Dataset notes

Sample data for 1 tasks

Challenge setup

Evaluation will assess the LlamaIndex agent's accuracy in financial modeling, the completeness and clarity of its audit trails, and its ability to handle structured financial data.

Challenge input

{'financial_data_csv': 'string (CSV content)', 'model_parameters': 'dict'}

Agent execution

No agent tool is listed for this challenge.

Evaluated output

{'financial_model_output': 'dict', 'audit_trail': 'list of strings', 'summary': 'string'}

Evaluation checks
  • The generated financial projections must be numerically accurate based on input data and parameters.
  • The 'audit_trail' list must contain at least 3 distinct steps.
  • The average length of individual entries in the 'audit_trail' (in characters). • target: 50 • range: 20-100
Configured evaluation evidence
  • Audit Trail Detail target: 50
  • 1 public reference case
Configured runtime
  • Python execution harness

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

No tool records are attached.

Evaluation contract

  • Correct Model Calculation · Weight 1
  • Audit Trail Presence · Weight 1
  • Audit Trail Detail · Weight 1

Metrics

  • Audit Trail Detail · Target 50

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: {'financial_data_csv': 'string (CSV content)', 'model_parameters': 'dict'}

    Output: {'financial_model_output': 'dict', 'audit_trail': 'list of strings', 'summary': 'string'}

    Execution harness: Python

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Agent for Auditable Financial Model Generation