UHIS Support Desk & Bangla AI Assistant — Bangladesh
Live multi-channel support desk and bilingual AI assistant for SPICE users in Bangladesh
The Challenge
Health workers using the SPICE platform in Bangladesh's UHIS programme needed a reliable way to get help, and field support teams needed one place to receive, prioritise, escalate and resolve issues arriving from many channels, in Bangla as well as English.
The Approach
I built a multi-channel support desk and an AI support assistant for Medtronic LABS' Bangladesh field operations. Issues arrive by web form, WhatsApp, an embeddable widget, email, API or Telegram and land in one queue with SLA tiers (P1–P5 plus outage), breach timers, auto-assignment to the least-loaded agent, and an escalation matrix that follows the country's division and district structure across four support tiers (L1–L4). Alongside it, a RAG-based WhatsApp assistant answers how-to questions from the product knowledge base, detects Bangla and "Banglish", replies in the user's language, transcribes voice notes, and opens a structured ticket when it can't resolve an issue.
The Result
The support portal is live at support.medtroniclabs.org, giving field teams a single, auditable view of every issue, with region-scoped access, CSAT feedback over WhatsApp, a public help centre, and analytics dashboards with export for programme reporting. The AI assistant architecture first built for Sierra Leone now serves a second country in a second language.
Field Lessons
Support design follows the shape of the health system
Escalation only works when it mirrors how the programme is actually run. Mapping support tiers to divisions and districts made ownership obvious instead of debated.
Meet users in their language and their channel
Health workers already live on WhatsApp and think in Bangla. Language detection and voice notes removed barriers that no amount of training would.
AI should hand off, not dead-end
When the assistant can't answer, it creates a structured ticket with context, so the human agent starts from the facts instead of from zero.
Deliverables
- Multi-channel ticket intake (web, WhatsApp, widget, email, API, Telegram)
- SLA tiers with breach timers and least-loaded auto-assignment
- Country escalation matrix across L1–L4 support tiers
- Bangla/English RAG WhatsApp assistant with voice-note support
- CSAT, public help centre and analytics dashboards with CSV/Excel export
- Region-scoped role-based access and audit logs
Tools Used
Want similar results for your organization?
Let's discuss how I can help improve your workflows, build scalable systems, and coordinate AI tools that deliver real operational impact.