Initiative overview
Documents are the backbone of governance, commerce, and social life. Yet virtually all leading Document AI systems, designed for high-resource languages like English, perform poorly on Khmer, Cambodia’s official script used by over 17 million people. This creates a compounding digital divide: government officials cannot efficiently query Khmer-language policy archives; businesses struggle to automate Khmer invoice processing; and citizens face barriers accessing public information in their own language. No commercially available AI solution adequately addresses this gap.
Government Use Cases in Action:
1. Admin Assistant: Summarises official letters, retrieves staff records, and answers queries about internal reports in real time.
2. Policy Assistant: Enables officials to query national AI strategies (e.g., NAIS), MEF IT frameworks, and regulatory documents conversationally.
3. Data Entry Automation: Extracts structured data from invoices, sale reports, and forms, dramatically reducing manual processing time.
KIDAI-A locally hosts open-source LLMs (Qwen, Gemma, GPT-OSS variants) to ensure data sovereignty, no sensitive government data leaves Cambodia’s infrastructure. The platform is offered through Subscription (PaaS), API services (OCR, STT, MT, RAG), Knowledge Hub digitisation services, and vertical solutions. Target customers include government agencies, education and research institutes, the financial sector, enterprises, SMEs, and legal services. The team is pursuing co-development partnerships with government ministries and plans to expand KIDAI-A into a national-grade AI infrastructure for Cambodia’s digital government ecosystem. KIDAI-A represents more than a product, it is a national movement. Cambodia, like many lower-middle-income nations, faces a structural risk: dependence on foreign AI tools that neither understand Khmer culture, nor protect Cambodian data. KIDAI-A is Cambodia’s answer to this challenge, a fully sovereign AI stack designed, trained, and deployed by Cambodians, for Cambodians.
Results, outcomes and impacts: KIDAI-A is actively under development with the following milestones achieved and projected:
1. Proprietary Khmer speech recognition model trained on 500+ hours of audio data, achieving high transcription accuracy for Cambodian speech.
2. In-house OCR and document layout analysis models functional for Khmer/English document extraction.
3. Working prototype deployed locally — vector store loaded with 406+ government documents, demonstrating real-time multilingual Q&A.
4. Projected impact: reduce document processing time for government staff by 60-80%; enable digital access to Khmer-language policy archives; support Cambodia’s national digitisation agenda.
