In Sri Lankan financial institutions, verifying customer identity cards (NICs), proof-of-billing, and multi-page bank statements has traditionally been an excruciating manual bottleneck. Employees spend 20 to 30 minutes cross-checking numbers, deciphering skew angles, and manually transcribing text.
Solving the Low-Resource Language Moat
Off-the-shelf vision OCR models routinely fail on Sinhala and Tamil characters or blurry mobile photographs. With DocScan, we engineered a dedicated multi-stage pipeline combining image de-skewing, perspective correction, and fine-tuned multilingual Vision-Language Models. The result: end-to-end customer verification slashed from 30 minutes to under 5 minutes.
