DocAI

May 10, 2024

• DocAI varifies and extracts document data used for app registration.
• It works for English and Bangla language and able to detect different types of varification documents.
• The core model is based on Donut model.
• It is further trained on synthetic data and available limited real data. The synthetic data was generated with SynthDog and DocSim.
• The system is served using Nvidia-Triton to maximise GPU utilization and ensure scalability which enables real-time usage.