Page 96 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

usability, and effectiveness of the chatbot [4][5]. The goal was to continuously improve the
system by integrating feedback from internal teams, target users, and broader deployment
settings. The initial testing layer, considered a quality control test, focused on generating a high-
quality multilingual corpus derived from health topics identified through FGDs. This database
was enriched with data from trusted sources like WHO, IFRC, and UNICEF. The curated corpus
of 165 common health queries was stored in a vector database and tested for accuracy, clarity,
and cultural relevance. Responses were refined to remove technical jargon, US-centric data,
and any diagnostic content. Subsequently, two phases of user testing were conducted: (1)
A workshop with 6 community members tested the Telegram interface, conversational flow,
and language handling. Feedback was gathered via FGDs. (2) 20 participants from Lebanon
(CHWs and LRC/IFRC volunteers) tested the WhatsApp version over five days [6]. Feedback was
collected using the METABase dashboard and reflective journaling. The third layer of testing
included 50 participants from vulnerable communities in Lebanon and Jordan who tested HIBA
for two weeks. Post-trial assessments via phone interviews and pre-defined metrics provided
both qualitative and quantitative feedback on accessibility, clarity, and user trust. Finally, a
social marketing campaign deployed HIBA in a Lebanese city with a diverse urban setting.
Materials tailored to subgroups (e.g., students, parents) were disseminated via NGOs, schools,
and parishes using QR-coded posters, flyers, and videos. Recently, the campaign extended
to displacement centers housing war-affected populations. The campaign’s reach and impact
were reinforced by a community mobilizer and evaluated through performance monitoring
on the chatbot dashboard. Overall, the chatbot responses were assessed using a structured
rubric and Likert scale, evaluating clarity, appropriateness, and trustworthiness. Commonly
asked topics and emergent patterns were also tracked. The expected impact includes increased
health literacy, reduced misinformation, and improved community engagement in health
decision-making, especially for populations that typically fall outside the reach of conventional
health promotion strategies [11][13][14].

Use Case Status: Active, with ongoing deployment and iterative refinement�

Partners:

• Lebanese Red Cross (LRC)—Youth Sector [15],
• Jordan National Red Crescent Society (JNRCS) [16]
• IFRC MENA Regional Office [17]

2�2 Benefits of use case

HIBA ensures equitable access to health information, particularly for marginalized populations.
By providing accurate, evidence-based health messaging, the chatbot empowers users with
preventive healthcare knowledge, reducing reliance on overburdened healthcare systems. It
also facilitates early identification of health concerns, enabling timely medical interventions.

HIBA functions as an AI-driven educational tool for health literacy. By using interactive Q&A,
multilingual content, and speech-based engagement, it ensures that people with low literacy
levels or disabilities can access relevant, understandable health information. Additionally, it
supports continuous learning for community health workers and volunteers.

By targeting underserved populations in LMICs, HIBA reduces health disparities caused
by socioeconomic and linguistic barriers. The AI-driven chatbot removes stigma, enabling
individuals - especially women and refugees - to access trusted health guidance anonymously

60