Call for proposals: AI-enabled multichannel hotline for LMAC

الوصف

The Lebanon Mine Action Authority (LMAA), through the Lebanon Mine Action Center (LMAC), leads and coordinates mine action in Lebanon, including explosive ordnance risk education and public communication. Members of the public currently reach the LMAC mainly through a phone hotline, and the full burden of answering calls falls on LMAC officers. LMAC intends to extend this hotline with a multichannel, multilingual AI-enabled support system. The service is intended to strengthen the hotline, not to replace it. The assignment is facilitated by HAMAP-Humanitaire under Project BRIDGE, supported by the Belgian Federal Public Service, Peacebuilding Unit. ## **Purpose of the assignment** Design, build, test, deploy and hand over a multichannel hotline support system: at least one digital messaging channel (WhatsApp, social media messaging, web chat or another agreed channel) and, subject to feasibility, a voice component in the form of structured voice-message intake or automated voice response. Languages: Arabic as the primary language, with understanding of Lebanese Arabic and answers in formal Arabic, English, and French as an option. ## **Critical design constraints, non-negotiable** - **Controlled content only**: no open generative AI. All responses must be based on a knowledge base and decision tree approved by LMAC. - **Mandatory fallback and escalation**: any urgent, sensitive, unclear or out-of-scope enquiry is immediately directed to the hotline, a human operator or an officer callback. - **Arabic language quality**: performance in Arabic, and in Lebanese Arabic in particular, must be evidenced and validated during the inception phase, before the platform choice is confirmed. - **Independence after handover**: content and contact details must be updatable by LMAC staff without vendor intervention and without vendor lock-in. - **Predictable running costs**: target below USD 200 per month, ideally below USD 50 for the pilot scope. - **Data protection**: GDPR p