Page 29 - AI Ready – Analysis Towards a Standardized Readiness Framework

AI Ready – Analysis Towards a Standardized Readiness Framework

4�4�1 Neak Pean HealthTech - Khmer Telemedicine Chatbot

This use case [2] solves the problem of access to medical care, inefficient management of
waiting times, and long queues. Key solutions include speech-speech local language Chatbot,
medical records, pre-health assessment, summary report generation, doctor dashboard, using
a mobile-based solution. Data includes audio speech data, (history recording in audio and later
converted to text). The Models include Khmer ASR (speech engine), Text-to-speech (TTS) (fast
speech), and chatbot (sentenceBERT for finetuning), for accent handling more data collection is
needed, currently, this model is based in the central City. Pilot deployments include deployment
in the Partnership Ministry of Post and telecommunication and publications for Khmer ASR are
available.

4�4�2 Improving Early Detection of Neonatal Asphyxia with Smartphone-
based AI Technologies

This use case [68] [69] aims to develop more robust and reliable AI-based solutions for
neonatal asphyxia detection on smartphones. This medical condition is critical for newborns
who experience oxygen deprivation during birth. The use case uses data from more than 1000
one-second-cry voice samples from existing datasets. The regression and ensemble model
applied in this use case makes it amenable to retraining these models based on the regional-
specific data so that the machine learning result is adaptable to regional problems.

The model is deployed in smartphones that are affordable and ubiquitous for most of the
population, making the solution deployable in resource-limited settings. So far, 4 pilots in 4
hospitals from different divisions have been set up for a year.

Collaboration with hospitals to collect data on a larger and more diverse population of
newborns, including data from different ethnicities, gestational ages, and birth complications
can further make the models generalizable. Comprehensive data collection would also include
collecting data from pre-birth stages (fetal heart rate, maternal health data) to identify potential
risk factors. Simulating variations in crying sounds due to background noise, microphone quality,
and different recording environments would make the models more reliable and robust.

4.5 Public Services

AI technologies could be integrated into different public services to address critical requirements
such as access to precise policy information, complex societal issues, the need for contextualized
local engagement. Technologies such as large language model, generative AI, text detection
and other AI technologies are applied in these use cases, demonstrating the power of AI
technology in the domain of public services.

4�5�1 Enhancing Transparency and Accountability in Public Procurement and
Project Monitoring

This use case [57][58] solves the problem of flagging corruption and irregularities in the
procurement process with the help of an AI model trained on tender documents. The consumer
of inference is the Prevention and Combating of Corruption Bureau (PCCB), a government
oversight body. The data used in this use case is text data, coming from historical records

22