An AI Translation App That Understands the Conversation, Not Just the Message

In a Nutshell

Challenge: Over a billion people use WhatsApp across 60+ languages, but there’s no native way to follow a conversation in a language you don’t speak. Existing workarounds - copy-pasting, tap-to-translate - are tedious, strip context, and can’t keep up with fast-moving chats.

Solution: A companion app that syncs WhatsApp chats with full, real-time translation. The AI uses conversation history to translate accurately, not just literally. Messages arrive on the other side as normal WhatsApp messages - no indication translation is involved.

Results:

4.7 App Store rating
Users across 40+ countries
Strong month-over-month growth
Sub-100ms response times on >200k messages/day, 99.9% uptime
60% reduction in per-message cost through intelligent architecture

Services: Product Strategy · AI Architecture · Full-Stack Engineering · UI/UX Design

The Challenge: Fragmented Financial Data Across Operations

Ruggable operates a complex global business with multiple production facilities and subsidiaries across different countries. The accounting team faced a month-end bottleneck, waiting at least one week after period close before they could begin their general ledger work.

Financial information was scattered across multiple data sources: five different payment service providers (PayPal, Klarna, Afterpay, Stripe, Shopify, and Amazon), production systems, shipment tracking, returns processing, and deposit and refund workflows.

Each source had different data structures, transaction naming conventions, and status definitions, making reconciliation a manual, time-intensive process that limited executives' access to timely financial insights.

Why WhatsApp's Built-In Translation Falls Short: AI Translation quality improved. Multilingual chat didn’t.

Over a billion people use WhatsApp across 60+ languages. But if you join a chat in a language you don't speak, you're stuck copy-pasting or tapping to translate one message at a time.

These methods are tedious, they strip context, and in fast-moving or dense chats- whether that’s for business, group chats, or personal chats - you just can’t keep up.  

While translation quality had improved with LLMs, no one was turning these powerful models into an invisible middle layer, in the place where conversations actually happen. Our vision: Build full-chat translation that feels native to both sides of the conversation.

Real-Time Chat Mirroring with Invisible Translation

WhatLingo works alongside WhatsApp as a companion app to create a mirrored view of the selected WhatsApp chats. Users see everything they need on WhatLingo, reply in their own language, and messages are still sent and received through WhatsApp as normal.

For the user:

Instant full chat translation so users are never behind
Write outgoing messages and have them translated on send
Voice messages are transcribed and translated

For their contacts:

Messages arrive through WhatsApp exactly as normal
No indication translation is involved

WhatLingo Key Results and Metrics

WhatsApp has no native full-chat translation. WhatLingo solves this with a companion app that translates entire conversations in real-time — both incoming and outgoing — without the other person ever knowing.

Full chat translation that lets users keep up with fast-moving conversations instead of falling behind
Invisible to contacts: messages appear as normal WhatsApp messages
Higher-quality translations through contextual awareness — the AI understands the conversation, not just individual messages
4.7 App Store rating, strong month-over-month growth, users across 40+ countries

Product Strategy: Speed Over Control in Multilingual Chat

Early on, we made a deliberate decision on the user experience: Prioritize speed and flow over control and learning.

Most translation tools assume users want to review and edit each message. We bet the opposite- that users would trust high-quality translation if it was consistent and didn't interrupt them. The first goal wasn’t helping users learn the language. It was making sure they could keep up.

That choice shaped everything: how the experience works, how translation is applied, and which features we intentionally left out. The result is an experience where translation disappears. Users stop thinking about it — they just have the conversation.

Context-Aware AI Translation Using Conversation History

"He missed it" translates differently depending on whether someone missed a goal or misses a teddy bear. Without conversation history, translation is guesswork.

WhatLingo sends the previous three messages as context with each translation request. The LLM understands the conversation — tone, subject, pronouns — not just isolated text. Users can optionally share grammatical gender for more accurate outgoing messages.

Multi-LLM Architecture for Reliable, Private, Low-Cost Translation at Scale

At scale, every millisecond and every cent matters. Our architecture balances quality, cost, and uptime across multiple LLM providers. 

Dynamic model selection: The system routes translation requests based on cost and real-time provider availability — not a single locked-in model.
Fallback Architecture: Automatic failover between providers ensures 99.9% uptime even during individual API outages.
Edge Processing: Cloudflare Workers enable sub-100ms response times by processing requests at the network edge.
Cost Optimization: Intelligent caching and model tiering reduced per-message costs by 60% without quality degradation.
Privacy by design: Messages aren't stored. No user profiling, no persistent identifiers. The backend orchestrates AI providers — it doesn't become one

Whitespectre is an international team. We live this problem daily—keeping up with the football group chat in Catalan, parsing voice notes from the plumber who assumes our French is better than it is, coordinating with in-laws across time zones in languages we're still learning.

When ChatGPT and the other LLMs launched, translation quality improved. But the workflow stayed broken. Still copy and paste or tap to translate. We saw the gap: no one was turning these tools into an invisible but powerful middle layer, in the place where conversations actually happen.

Allison KellmanCPO Whitespectre

Overview - At a Glance

WhatsApp has no native full-chat translation. WhatLingo solves this with a companion app that translates entire conversations in real-time—both incoming and outgoing—without the other person ever knowing. 

Full chat translation that lets users keep up with fast-moving conversations instead of falling behind
Invisible to contacts: messages appear as normal WhatsApp messages
Higher-quality translations through contextual awareness—the AI understands the conversation, not just individual messages
4.7 App Store rating, strong month-over-month growth, users across 40+ countries