Module: Legion::LLM::Inference::Steps::Classification

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

Constant Summary collapse

LEVELS =
%i[public internal confidential restricted].freeze
PII_PATTERNS_CORE =
{
  ssn:   /\b\d{3}-\d{2}-\d{4}\b/,
  phone: /\b(?:\+?1[\s.-]?)?\(?\d{3}\)?[\s.-]?\d{3}[\s.-]?\d{4}\b/
}.freeze
EMAIL_PATTERN =
/\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}\b/
PII_PATTERNS_EXTENDED =
{
  ip_address:     /\b(?:\d{1,3}\.){3}\d{1,3}\b/,
  date_of_birth:  %r{\b(?:0[1-9]|1[0-2])[/-](?:0[1-9]|[12]\d|3[01])[/-](?:19|20)\d{2}\b},
  zip_code:       /\b\d{5}(?:-\d{4})?\b/,
  mrn:            /\b(?:MRN|mrn)[:\s#]?\s?\d{4,12}\b/,
  account_number: /\b(?:account|acct)[:\s#]?\s?\d{6,20}\b/i,
  license_number: /\b[A-Z]\d{3,8}\b/,
  url:            %r{\bhttps?://[^\s<>"']+}i,
  vin:            /\b[A-HJ-NPR-Z0-9]{17}\b/,
  npi_number:     /\b\d{10}\b/
}.freeze
PII_PATTERNS =
PII_PATTERNS_CORE.merge(email: EMAIL_PATTERN).merge(PII_PATTERNS_EXTENDED).freeze
PHI_KEYWORDS =
%w[
  patient diagnosis medication prescription
  dob date-of-birth mrn medical-record
  npi clinical treatment health-plan
  beneficiary insurance-id group-number
  lab-result radiology pathology
  admission discharge procedure
].freeze

Instance Method Summary collapse

Instance Method Details

#step_classificationObject



46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
# File 'lib/legion/llm/inference/steps/classification.rb', line 46

def step_classification
  classification = @request.classification || compliance_classification_default || default_classification
  unless classification_enabled?(classification)
    log_step_debug(:classification, :skipped, reason: :disabled)
    return
  end

  declared_level = Legion::LLM::Settings.config_value(classification, :level)
  log_step_debug(:classification, :scan_start, declared_level: declared_level)
  scan            = scan_content_for_sensitive_data
  effective_level = upgrade_if_needed(declared_level, scan)
  upgraded        = effective_level != declared_level

  redact_sensitive_content(scan)
  enforce_phi_cloud_gate(effective_level)
  log_step_info(
    :classification,
    :scan_complete,
    declared_level:  declared_level,
    effective_level: effective_level,
    contains_pii:    scan[:contains_pii],
    contains_phi:    scan[:contains_phi],
    pattern_count:   scan[:patterns].size,
    upgraded:        upgraded
  )

  @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