Khmer eKYC OCR API uses AI to instantly extract data from Cambodian identity documents in Khmer and English, and already powers CamDigiKey, Cambodia's official government single sign-on application. Available completely free via CamDX to all 45+ registered public and private members, including banks, insurance companies, and government agencies, it requires no integration cost and is open to both public and private CamDX members. It has processed more than Khmer ID card images, as of now.
Initiative overview
Background and problem
Reliable Optical Character Recognition (OCR) for Khmer-script documents has historically been unavailable, creating a significant bottleneck in Cambodia's digital transformation. Financial institutions, public agencies, and businesses were forced to rely on manual data entry for identity verification, a slow, error-prone, and costly process. TSC's response to this gap, the Khmer eKYC OCR API, earned the Best R&D Award 2023 from the Ministry of Post and Telecommunication (MPTC) of Cambodia, recognising it as the country's most impactful research and development achievement of the year.
What the initiative does
TSC's AI team built the eKYC OCR API from the ground up using deep learning models, extracting structured data from Cambodian National IDs (Khmer and English), Passports (MRZ and biographical), and Driving Licences. The API is available entirely free of charge via CamDX, Cambodia's national data exchange platform, to all 45+ registered members, including banks, insurance companies, microfinance institutions, and government bodies. Both public- and private-sector CamDX members can access it freely. Notably, it powers CamDigiKey, Cambodia's official government single sign-on (SSO) app, enabling citizens to authenticate and access public digital services such as online business registration using their national identity documents.
Objectives
- Eliminate manual identity data entry;
- accelerate eKYC compliance for NBC-regulated institutions;
- deliver inclusive digital services in Khmer;
- provide a sovereign, locally-hosted alternative to foreign OCR;
- and democratise access by offering the API free of charge through CamDX, removing cost as a barrier to digital transformation for all member organisations.
Future evolution
Having surpassed 8 million processed requests (Khmer ID card images or passport images) with zero cost to integrators, TSC will continue expanding document coverage, growing the CamDX member base, and deepening integration with Cambodia's national digital identity infrastructure. Provincial government adoption and ASEAN-level replication are key targets for 2026 and beyond.
Other relevant details
Results, outcomes and impacts: The initiative was recognised with the Best R&D Award 2023 of Cambodia, validating the initiative's technical innovation and public impact. The API powers CamDigiKey, Cambodia's official government single sign-on (SSO) application, and is accessible for free to all 45+ CamDX members (banks, insurance companies, microfinance institutions, and government agencies). It has processed more than 8 million requests at no cost to any integrating organisation. Accuracy rates for National ID extraction exceed 95%. Identity verification time has dropped from several minutes to under 5 seconds per document, account-opening costs have measurably fallen, and human data-entry errors have been eliminated in integrated workflows.
Challenges and lessons learned: Challenges: Khmer script's complex orthography, including stacked consonants, diacritics, and multi-level vowel placement, made standard OCR approaches insufficient. Acquiring diverse, labelled training data for Cambodian identity documents was difficult due to privacy constraints. Worn or low-quality document images also posed accuracy challenges. Responses: TSC built a custom AI-based pipeline trained specifically on Khmer script. A secure, privacy-preserving data collection protocol was established with government partners. Image pre-processing pipelines were added to handle poor lighting, skew, and noise. Lessons learned: Sovereign AI capability requires long-term investment in local training data and domain-specific models. Open API design enables faster ecosystem adoption than proprietary integrations. Success conditions: Institutional commitment, access to quality training data, regulatory alignment, and a robust digital infrastructure are all prerequisites.
