Wongreang

The Indramayu Regency government runs an internal app called Wongreang for document management and inter-agency coordination. But input was still manual: staff read physical documents and retyped them into the system. Old archives (pre-2022) existed only as image scans — not searchable. Staff spent 3–5 hours a day on transcription alone, and searching historical documents could take half a day.

Indonesian OCR on government documents has 3 specific difficulties: (a) a mix of printed text (decrees, reports) and handwriting (dispositions, leadership notes) — requiring a dual pipeline; (b) widely varying scan quality — from modern office scanners to tilted phone photos; (c) administrative terminology (register numbers, letter numbers, Indonesian date formats) needs dedicated post-processing for accurate metadata extraction.

We didn't replace the Wongreang app — we just added an AI layer via API. The Python pipeline: PaddleOCR for print + a Tesseract fallback for handwriting, deskew/denoise pre-processing, regex post-processing to extract letter numbers and dates, and document classification via a lightweight classifier. Results are pushed to Wongreang's Laravel API and indexed into Meilisearch + pgvector embeddings for semantic search. Staff simply upload a scan → the document appears in the system with metadata filled in, ready to review and approve.

Pilot live across several Indramayu Regency offices within 4 weeks. Manual transcription dropped significantly — staff now focus on verification and decisions instead of retyping. Years of backlogged archives are being backfilled, and document searches went from hours to seconds. Currently rolling out to more offices, plus expanding into summarization (condensing long reports) and content-based disposition routing.