Polyglot Translate Cloud is a self-learning translation platform built and operated by Elevio Solutions. It translates website content through a three-engine machine translation cascade, stores every approved segment in a Translation Memory, and serves repeated segments at no additional cost. The platform is live in production and powers the Polyglot Translate WordPress plugin.
The Challenge
Commercial website translation has a pricing problem. Most services charge per word or per credit, at the same rate every time, even when the sentence being translated is one the system has already processed a hundred times. On a WooCommerce store where thousands of products share attribute names and template copy, the invoice keeps growing while the amount of genuinely new translation work shrinks. That struck us as a business model, not a technical necessity.
There was a second motivation closer to home. We build for clients in Serbia and across the region, and we wanted a platform that treats Serbian, Croatian, Bulgarian, and Slovenian as first-class languages rather than an afterthought behind the large Western European pairs. The only way to guarantee that was to build it ourselves.
The Solution
The core of Polyglot Translate Cloud is a Translation Memory that learns. Every approved segment is stored with a SHA-256 key, so an exact repeat is found with a single hash lookup. When the same segment appears again, the platform serves it straight from memory at no cost. You pay to translate a sentence once; after that, repeats are free, and the memory becomes more useful with every approval.
Segments that miss the memory go to a three-engine cascade: a self-hosted OPUS-MT engine, Gemini Flash-Lite, and a self-hosted Argos engine. Two of the three run on our own servers, which keeps per-segment costs under our control rather than a vendor’s.
Every machine translation enters the memory as a candidate. Review and approval move segments up through the trust tiers, from candidate to verified to gold, and the tier travels with the segment, so you always know whether a translation is raw machine output or something a person has signed off on. AI in the engine, humans at the wheel.
The platform covers more than 30 languages, with deliberate depth in the Southeast European pairs, and billing is self-serve through LemonSqueezy.
Technical Architecture
Polyglot Cloud is TypeScript from the API to the dashboard.
| Layer | Technology |
|---|---|
| API | Node.js 20+, TypeScript, Express 4 |
| Dashboard | Next.js 14 |
| Database | PostgreSQL 16 with Prisma (62 models) |
| Queues | Redis 7 with BullMQ |
| Translation engines | Self-hosted OPUS-MT, Gemini Flash-Lite, self-hosted Argos |
| Billing | LemonSqueezy |
| Infrastructure | DigitalOcean |
Long-running translation work is queued through BullMQ on Redis, so bulk jobs are processed in the background instead of inside a web request. The domain is mapped across 62 Prisma models on PostgreSQL 16, enough to treat memories, trust tiers, and billing state as real, queryable data rather than loose JSON.
Every release is checked automatically before it reaches a customer. For a system that stores client content and handles billing, we treat that as the baseline rather than an achievement.
What Makes It Different
- The economics improve with use. A per-word service bills the thousandth repeat of a string at the same rate as the first. Polyglot serves repeats from Translation Memory at no cost, so the effective price of maintaining a growing multilingual site falls as the memory fills.
- Provenance on every segment. The candidate, verified, and gold trust tiers record whether a translation is fresh machine output or human-approved, so quality is auditable instead of assumed.
- Self-hosted engines. OPUS-MT and Argos run on our own infrastructure. That is both a cost decision and a control decision.
- A real ecosystem around it. The platform powers the Polyglot Translate WordPress plugin, and the Polyglot Connector plugin is published on WordPress.org at version 1.0.0, built on the WordPress 7.0 Connectors API. That work draws on the same experience we sell through custom WordPress plugin development.
Status Today
Polyglot Translate Cloud is live in production on DigitalOcean, with a 72GB production database behind it. It is one of the products we build and operate alongside client work, and it exists partly for the reasons laid out in why we build our own SaaS products: a studio that ships and runs its own platform learns lessons no client project can teach.
It also serves as a working reference for what we build for others. The stack and the test discipline on this platform are the same ones we bring to SaaS and software development engagements.
If you are scoping a platform of your own, look through the rest of our work or tell us what you are building. You will get a straight answer on scope and cost from the people who wrote this one.