Classification & MIP Mapper Project

Fixed Price: $1,500 We are extending the existing Smart Stack 3.0 Trusted Data Refinery to strengthen enterprise classification and connect its results to customer-defined Microsoft Purview sensitivity labels. The completed workflow will allow Smart Stack 3.0 to detect sensitive information, map detected entity types to the customer’s Purview labels, and apply the appropriate label to supported files through a Linux-based Microsoft MIP implementation. We detect. We label. Purview enforces. Scope The developer will: Review the existing Presidio and Trusted Data Refinery architecture, APIs, schemas, tests, and sanitized outputs. Build five custom credential and infrastructure-secret classifications: Microsoft Entra Client Secret Azure DevOps Personal Access Token GitHub Personal Access Token Amazon S3 Secret Access Key Azure Storage Account Key Define patterns, context, confidence scoring, validation logic, and positive, negative, and ambiguous test cases. Measure precision, recall, F1, processing time, and throughput against the approved synthetic test set. Validate the supported Microsoft MIP SDK/API path in the approved Linux environment. Authenticate through the provided Entra service principal and retrieve customer-defined Purview sensitivity labels. Apply and programmatically verify classification-only labels on approved DOCX and PDF test files. Connect structured Refinery classifications to the MIP Mapper workflow. Support customer-specific classification-to-label mappings, priority rules, decision reasons, and audit results. Integrate the demonstration into the private Classification Lab and MIP Mapper Lab within the Smart Stack 3.0 Playground. Deliver source code, configuration templates, tests, setup instructions, technical findings, and handoff documentation through the designated GitHub and Azure DevOps repositories. Label Rules V1 will: Preserve an existing label of equal or higher sensitivity. Upgrade when the approved mapped label has higher sensitivity. Never downgrade automatically. Record every label decision and reason. Skip and report unsupported, ambiguous, or unauthorized actions. Architecture The work will extend the existing Smart Stack 3.0 architecture. It will not create a separate product, authentication system, database, backend platform, or production administration interface. The lightweight Playground UI will use existing Lovable design and API conventions. Backend classification and labeling services will remain outside the browser. Microsoft credentials will not be stored in Lovable or source code. Deliverables Architecture, environment, and implementation assessment. Five integrated custom classifications with automated tests. Accuracy and performance benchmark results. Validated Linux-based Microsoft labeling implementation. Retrieval of Purview sensitivity labels. Verified DOCX and PDF labeling. Connected classification-to-label mapping workflow. Functional Classification and MIP Mapper Labs. Complete source code, configuration, documentation, and technical handoff. Acceptance Criteria The project is complete when: All five classifications operate within the existing Refinery workflow. Each classification includes positive, negative, and ambiguous tests. Precision, recall, F1, processing time, and throughput are documented. The Microsoft labeling implementation runs in the approved Linux environment. Service-principal authentication and Purview label retrieval succeed. Approved DOCX and PDF files are labeled and verified. Classification results feed the mapping workflow. Label decisions are deterministic and auditable. Unsupported and failed actions are reported clearly. The Playground demonstration is functional. Jas can reproduce the workflow using the delivered code and documentation. All work product is committed to the designated repositories. Project Boundaries The engagement does not include production deployment, production customer data, encryption, rights protection, DLP, retention, deletion, access enforcement, file migration, Copilot integration, more than five new classifications, or rebuilding the broader Smart Stack application. Budget The total fixed price is $1,500, including implementation, testing, documentation, revisions, and technical handoff.