P2.4: Generate Structured & Scored Outputs

[Virgo-Alpha]

This ticket focuses on transforming the agent's final output from a simple block of text into a rich, structured, and more actionable format. Instead of just a text memo, the system will generate a full JSON object that can be easily rendered into a dashboard. This includes creating a scoring mechanism to provide at-a-glance insights into a potential investment.

Implementation Tips

• Pydantic Schemas: This is the core of the implementation. The JsonFormattingAgent will be given a Pydantic output_schema that defines the exact structure of the final report. This is the most reliable way to enforce a consistent JSON output from the LLM.
• Dashboard Summary: In your Django models (e.g., ScanJob), create a new JSONField called summary_data. After the full report is generated, a small utility function can parse it to extract key metrics (e.g., employee count, funding amount) and save them to this summary field for quick loading on a dashboard.
• Scoring Mechanism: Develop a simple scoring algorithm. This can be a Python function that takes the final JSON report as input. It can assign points based on certain criteria (e.g., +5 points if founders have previous exits, -10 points if financial analysis shows high burn rate). The final score can be saved to a new score field on the ScanJob model.
• Final Report Display: Create a new template and view that can take the final JSON object and render it as a clean, easy-to-read web page or dashboard, with different sections for each key in the JSON.

Acceptance Criteria (Checklist)

• A Pydantic model defining the full structure of the final report exists.
• The JsonFormattingAgent is configured to use this model as its output_schema.
• The final output from the agent workflow is a valid JSON object matching the schema.
• A simple scoring function is created and calculates a score based on the JSON report.
• The final score and a summary of key metrics are saved to the ScanJob database record.

Unit Tests

• Scoring Function Test: Write unit tests for your scoring algorithm. Feed it mock JSON reports with different data points and assert that it calculates the expected score for each case.
• Model Tests: Test that the summary_data and score fields on the ScanJob model can be saved and retrieved correctly.