Our client’s unstructured estate — thousands of SharePoint sites of varying ownership and activity, department drives, OneDrive, plus structured stores — cannot be mapped by interviews alone. This role stands up AI-enabled discovery and classification to find personal information, tag it to our client’s record types and populate the customer and associate data maps, with a human validation step at every stage. Phase 1 runs on a bounded dataset (and on our client’s existing Microsoft capability if external tooling has not yet cleared security review); Phase 2 scales across nine functions; Phases 3–4 hand over classification rules and re-scan to prove evidence traceability.
Nothing you build touches our client production customer or employee data until it has passed our client’s AI vendor/tool risk assessment and data processing and security review — you will help prepare that submission in week one. Output is never accepted on model confidence alone; the Data Engineer independently validates it.
What you will do, by phase
Phase / commitment | | What you will do | |
Phase 1 — Mobilize & Assess (wk 1–6) Full | | Prepare and submit the candidate tooling package (architecture, data flows, processing location, retention of scan results, access model) to the client’s InfoSec team for the AI vendor/tool risk assessment Deploy discovery tooling inside our client’s environment; run a bounded-dataset classification pilot (recommended: People & HR record types) and measure precision/recall against human review Define sensitive information types / classifiers for our client’s record taxonomy, including SPI and ePHI flags | |
Phase 2 — Execute in Waves (wk 3–16) Full | | Run discovery and classification across the remaining nine functions in waves aligned to the Compliance Business Analyst’s schedule Populate data maps: personal data elements, collection source, purpose, internal recipients/flows, external recipients, system location, record type, retention trigger/period Present AI output alongside independently validated findings; maintain a sampling-based validation log and error analysis Automate governance documentation where it accelerates the work (workbook population, lineage narratives) with review gates | |
Phase 3 — Governance & Operating Model (wk 5–18) Partial | | Hand over classification rules, classifiers and scan configurations with documentation so the client can operate them; define re-scan cadence for the health-check process | |
Phase 4 — Evidence & Audit Support (wk 17–28) Partial | | Run re-scan validation to confirm disposition took effect and no untracked copies remain; produce evidence traceability from scan → inventory → disposition record Answer auditor questions on how the inventory was produced and validated | |
Deliverables you own or co-own
Enterprise Data Inventory and Repository Register — AI discovery output presented alongside independently validated findings (co-owned with Data Engineer)
Customer and Associate Data Maps — populated for every record type carrying personal information (co-owned with Compliance Business Analyst)
Classification rule handover pack and re-scan evidence (Phases 3–4)
AI tooling risk-assessment submission and approval record
Must-have qualifications
5+ years applied machine learning, data science or automation engineering with hands-on data classification / sensitive-data discovery at enterprise scale
Production experience with at least one discovery/classification platform: Microsoft Purview (sensitive info types, trainable classifiers, DSPM), BigID, Varonis, OneTrust or equivalent
Designed human-in-the-loop validation with measurable quality (precision/recall, sampling plans, error analysis) and can explain results to non-technical reviewers
Python; Microsoft Graph / SharePoint APIs or equivalent for estate-scale scanning; secure handling of results inside a client tenant
Has prepared or contributed to an enterprise AI/security review (architecture, data flow, model/data residency) and shipped under its constraints
Understands personal information categories under CCPA/CPRA, SPI and ePHI well enough to design classifiers for them
Strongly preferred
Azure OpenAI / LLM-based document classification pipelines with evaluation harnesses; agentic workflows for governance documentation with review gates
BigID or Purview certifications; NIST AI RMF familiarity
Experience classifying legacy formats (mainframe extracts, COBOL copybooks, call-recording transcripts)
Prior privacy or records program where classification output fed retention labels or a data map
Tools and platforms
Microsoft Purview Information Protection / DSPM, BigID, SharePoint Online and Microsoft Graph, Azure services as approved by the client, Python, Excel function workbooks, Clarity PPM.
Never miss an opportunity! Create an alert based on the job you applied for.
