Module: Legion::LLM::Pipeline::Steps::Classification

Includes:
Legion::Logging::Helper
Included in:
Executor
Defined in:
lib/legion/llm/pipeline/steps/classification.rb

Constant Summary collapse

LEVELS =
%i[public internal confidential restricted].freeze
PII_PATTERNS =
{
  ssn:   /\b\d{3}-\d{2}-\d{4}\b/,
  email: /\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}\b/,
  phone: /\b(?:\+?1[\s.-]?)?\(?\d{3}\)?[\s.-]?\d{3}[\s.-]?\d{4}\b/
}.freeze
PHI_KEYWORDS =
%w[
  patient diagnosis medication prescription
  dob date-of-birth mrn medical-record
  npi clinical treatment
].freeze

Instance Method Summary collapse

Instance Method Details

#step_classificationObject



26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
# File 'lib/legion/llm/pipeline/steps/classification.rb', line 26

def step_classification
  return unless @request.classification || compliance_classification_default

  classification  = @request.classification || compliance_classification_default
  declared_level  = classification[:level]
  scan            = scan_content_for_sensitive_data
  effective_level = upgrade_if_needed(declared_level, scan)
  upgraded        = effective_level != declared_level

  @enrichments['classification:scan'] = {
    declared_level:    declared_level,
    effective_level:   effective_level,
    contains_pii:      scan[:contains_pii],
    contains_phi:      scan[:contains_phi],
    detected_patterns: scan[:patterns],
    upgraded:          upgraded,
    timestamp:         Time.now
  }

  if upgraded
    @warnings << "Classification upgraded from #{declared_level} to #{effective_level}: " \
                 "sensitive patterns detected (#{scan[:patterns].join(', ')})"
  end

  @audit[:'classification:scan'] = {
    outcome:     :success,
    detail:      "level=#{effective_level}, upgraded=#{upgraded}",
    data:        { declared: declared_level, effective: effective_level, patterns: scan[:patterns] },
    duration_ms: 0,
    timestamp:   Time.now
  }

  @timeline.record(
    category:  :audit, key: 'classification:scan',
    direction: :internal, detail: "level=#{effective_level}, upgraded=#{upgraded}",
    from:      'pipeline', to: 'classification'
  )
end