Skip to content
LuminForge
All case studies
Mobile · iOS & AndroidHealthcare · Mobile11 weeksIllustrative · Reference playbook

Helio Health — 24/7 care in eleven languages

Helio's call-center was the bottleneck: 14-minute hold times, six languages supported by human staff, no after-hours coverage. We shipped two native apps with an on-device voice interface and a fine-tuned medical chatbot — same clinical guardrails, eleven languages, zero hold time.

Disclosure

This brand is illustrative — not a paying client. The architecture, stack, and approach below are the real playbook we'd use to ship this engagement. Named-client case studies replace these as paying engagements ship.

4.8 ★
App Store rating
ENESहिन्दीالعربيةবাংলাاردو
50,000
Monthly active users
−42%
Support cost
11
Languages supported
320 ms
P95 response latency
01 · Challenge

What we walked into.

Helio served patients across South Asia and the Middle East with a five-language call center that couldn't scale. Compliance and clinical accuracy were non-negotiable. Existing chatbot vendors couldn't handle code-switching or low-resource languages — and couldn't be self-hosted.

02 · Approach

How we built it.

  • 01

    Built native iOS (SwiftUI) and Android (Jetpack Compose) apps in parallel, sharing a small Kotlin Multiplatform layer for the audio pipeline

  • 02

    Fine-tuned a 70B medical model on Helio's clinical guidelines + 18 months of de-identified support transcripts

  • 03

    Shipped on-device whisper-style speech-to-text so consultations work offline and audio never leaves the patient's phone

  • 04

    Streamed responses with sub-second first-token latency by routing through Vercel Edge functions colocated with the user

  • 05

    Implemented a clinician escalation handoff — if the model isn't 95% confident, a human picks up mid-conversation with full context

03 · Results

What changed for the business.

50,000 MAU within six months; 4.8★ on both stores

Support costs reduced 42% in the first quarter after launch

Languages expanded from 5 → 11, including Sinhala, Pashto, and Bengali dialects

P95 latency held at 320 ms across regions thanks to edge-routed inference

Stack

Built with.

Swift / SwiftUIKotlin / ComposeKMPClaude (fine-tuned)WhisperVercel Edge
"The app handles 70% of the conversations our call-center used to take. The other 30% reach a clinician with the full transcript already summarized. It's a different operation."
DS
Dr. Sofia Alvarez
Head of Product, Helio Health
Now booking new projects

Have a project in mind? Let's talk.

Tell us what you're building. We'll come back with a concrete plan, timeline, and price — and an honest take on where AI gives you the most leverage.

  • Free 30-min discovery call
  • Concrete roadmap & pricing within 48 hours
  • No NDA needed for the first chat
Email hello@luminforge.ai

Typical response time: under 4 hours.