santhosh790/ecomsenseai

Purpose and entry point

ecomSenseAI is a Streamlit-based document-processing application that extracts vegetable items and quantities from PDF, image, and Excel orders, lets users validate results, and saves date-scoped order outputs. The composition/UI entry point is ecomSenseAI.py; domain and service layers support its upload-to-confirm workflow.

Domain model and catalog

Extracted items are represented as typed domain records, including detection status and confidence, in domain/models.py. Vegetable normalization and lookup are driven by catalog/alias data in data/vegetables.json and data/aliases.json, accessed through infrastructure/vegetable_catalog_repository.py. This catalog boundary separates source-document wording, OCR noise, and canonical output names.

Extraction pipeline

application/extraction_service.py is the central parsing and decision layer. It selects/handles document text formats, identifies item/quantity candidates, performs fuzzy catalog matching, and applies confidence/status policy. It produces domain detections rather than coupling parser logic directly to the Streamlit UI. Client- or supplier-specific interpretation is supported through data such as data/clients.json.

Document and OCR ingress

Document-specific text acquisition is isolated from extraction rules. infrastructure/document_readers.py handles PDF/Excel-style input reading, while infrastructure/ocr_engine.py handles image preprocessing and OCR. The OCR engine prefers PaddleOCR with OpenCV preprocessing and falls back to Tesseract; Tesseract additionally requires a host-installed binary. Extracted text is returned to the application layer for parsing and user preview.

Validation and order lifecycle

The primary control flow is: upload → read/OCR → extract detections → user validation/editing → confirm → persist and optionally publish. ecomSenseAI.py presents confidence controls and review states so low-confidence detections remain human-verifiable. application/reporting_service.py forms individual and consolidated order views, while application/vegetable_detection_service.py supports detection-oriented application operations.

Persistence, reporting, and saved-order boundaries

Confirmed rows are saved locally as dated CSV outputs and can be reloaded/revalidated by date and source file. infrastructure/persistence_service.py owns this storage boundary; infrastructure/assets_service.py supports stored upload/original-asset handling. Consolidation aggregates quantities across confirmed files for a selected date rather than treating each upload as an isolated final report.

External integrations and deployment runtime

Google Sheets publishing is optional and isolated in infrastructure/google_sheets_service.py, using Streamlit secrets for a GCP service account and target worksheet. Address-related data/services are separate (infrastructure/address_service.py, data/addresses.json), likely supporting order/report organization. Python dependencies are declared in requirements.txt; deployment-level system OCR/font requirements are captured in packages.txt, with Streamlit Cloud as the documented hosting target.