Feature/evaluation - Dataset Interface

[RonShakutai]

Summary

Adds real dataset loading capabilities to the evaluation flow, replacing the static mock-only UI with an interactive dataset interface. Users can load CSV/JSON files from local paths, preview records, and configure detection options based on dataset content. Aligns the entity schema with Presidio Analyzer's RecognizerResult format.

What's Included

Dataset Loading (Backend)

• New router backend/routers/upload.py — POST /api/datasets/load accepts a local file path, format (CSV/JSON), and column names. Parses the file, extracts entities if present, and stores records in memory.
  • CSV parser via csv.DictReader with configurable text/entities columns
  • JSON parser supporting both JSON arrays and JSONL format
  • Preview endpoint (GET /api/datasets/{id}/preview?limit=5) and full records endpoint
• New models — DatasetLoadRequest (path, format, text_column, entities_column) and UploadedDataset (id, filename, format, record_count, has_entities, columns)
• Sample dataset — data/sample_medical_records.csv with 10 medical records containing pre-tagged entities (PERSON, DATE_TIME, EMAIL_ADDRESS, PHONE_NUMBER, US_SSN, CREDIT_CARD, etc.)

Dataset Selector (Frontend — Setup Page)

• Replaced static dataset list with a dropdown selector containing a seed example and an "Add new dataset…" option
• Inline form for loading datasets: absolute path input, format dropdown (CSV/JSON), text column name, optional entities column name
• Selected dataset displays a green summary card with record count, format, entity status, and a preview of the first few records
• Detection options (Run Presidio / Run LLM) appear when a dataset is selected

Greyed-Out Sections

• Compliance Context and Data Access Constraints cards are visible but disabled (opacity-50, pointer-events-none) with "Coming soon" badges
• Data Access Constraints text notes that only cloud-based LLM processing is currently supported

Entity Schema Alignment

• Renamed Entity.type → Entity.entity_type across the entire codebase (backend models, mock data, frontend types, all components) to match Presidio Analyzer's RecognizerResult output format
• Added dataset_entities field to Record model for entities loaded from datasets

Conditional Flow

• Anonymization page — reads setupConfig from session storage and conditionally renders Presidio/LLM processing cards based on user selection
• Human Review page — "Skip Tagging" button for datasets with pre-existing entities; uses entity_type field throughout
• EntityComparison component — supports optional datasetEntities prop

screenshots