What does Presidio do and where in the pipeline should it sit?

Category: Security & Compliance

Answer

Presidio is a PII recognizer/anonymizer using spaCy + custom recognizers. It labels tokens with entity types (EMAIL, PHONE, SSN) and applies operators (redact, hash, mask). It sits BEFORE any LLM call so PII never reaches the model — that’s also where you keep PII for compliance.

Concrete examples from the fca project context

Example 1

AnalyzerEngine loads en_core_web_lg, plus custom recognizers for account numbers.

Example 2

AnonymizerEngine applies redact/hash/mask per entity type.

Example 3

In the LangGraph workflow the security_service node runs first, before the LLM call.

Data flow / flow chart

user_msg -> Presidio anonymize -> scrubbed_msg -> LLM call
  PII stored separately in audit log if compliance needs it

Takeaway

PII redaction belongs in the security node, BEFORE the LLM call, not after.