THE PROBLEM
Why this system existed
Audit teams need fast evidence discovery without weakening independence, confidentiality, review hierarchy, or traceability.
RAG · Audit Intelligence · Agents
Professional Services Audit Intelligence Platform
A retrieval-backed audit assistant with engagement-scoped retrieval, evidence-grounded answers, ranked risk signals, mandatory reviewer approval, and measurable quality gates.
MY ROLEAgentic workflow design, retrieval, anomaly detection, evaluation, and delivery
SCOPEAn audit intelligence workflow for evidence gathering, workpaper linkage, risk flagging, anomaly detection, and grounded standards research. Internal client documents are not exposed.
EVIDENCEProfessional experience · public-safe summary
34%less manual review
<4 minevidence search
28%less initial analysis time
OUTCOME
What changed
Reduced manual document-review effort by 34% and reduced evidence-search time from about 40 minutes to under 4 minutes per assertion.
REFERENCE ARCHITECTURE
Controls around the model
01Classify
02Route
03Retrieve
04Rerank
05Generate
06Validate
07Observe
DECISIONS
Trade-offs considered
Engagement-scoped indexes
Human review before client-facing output
Evidence-aware anomaly ranking
Immutable audit events and redaction
FAILURE CASE
What did not work
Cross-engagement retrieval and unsupported standards references were treated as release-blocking risks through scoped indexes and evaluation gates.
SECURITY BOUNDARY
Public-safe by design
This case study exposes patterns, not employer architecture. It uses synthetic data, no client identifiers, no internal prompts, no proprietary datasets, and no production endpoints.