Hi Gulrez! Cool project. Full disclosure: I work with JavaScript/TypeScript, so I don't know Python syntax, but strictly from an architectural logic perspective, this looks like a silent data contract mismatch inside call_classifier().
Since FastAPI heavily relies on Pydantic schemas for data validation (which acts very similarly to JSON Schema/AJV in the frontend world), you might want to double-check:
1. Response Serialization: Does the JSON structure returned by call_classifier() exactly match the fields and data types defined in the Pydantic model your orchestrator expects? If a field type doesn't match, or if an expected key is missing, Python might fall back to None (which serializes to null in JSON).
2. Strict Checks: Inside call_classifier(), ensure you are doing an explicit check on the parsed object (the Python equivalent of a strict === check, using if response_data is not None:).
Also, if your PDF generator is crashing, it's highly likely a domino effect because that null value is leaking deeper into the pipeline where a real string or object is expected.
Sometimes a fresh pair of eyes from a different stack helps! Wishing you the best of luck with debugging this!