Our strength is collaboration

For one of Poland's leading television networks, we engineered advanced HbbTV infrastructure that brings modern interactivity to traditional broadcast television.

The platform enables viewers to pause live broadcasts, restart programs from the beginning, and access catch-up content — all directly from their television set, without a separate streaming app.

The system is built on a custom API-driven architecture fully integrated with the broadcaster's existing broadcast infrastructure, bridging legacy television delivery with streaming-style interactivity.

We built and operate our own AI translation pipeline — not a wrapper around a third-party API, but a production system with structured JSON processing, multi-stage validation, and a persistent two-tier disk cache keyed by SHA-256 content hashes.

It handles UI keys, CMS content, and product catalogues while preserving placeholders, variables, and formatting integrity. The cache is atomic-write and content-hash-driven: when source content changes, the hash changes, the cache invalidates automatically — no manual purging.

The system powers translation across multiple live projects including TanosMotorsport (PL→EN/DE product pages via ISR) and Expedition Lapland (full multilingual static site). It runs as a self-hosted API or inline library — no vendor lock-in, full data ownership.

Squick solves a concrete problem: every time a developer opens a new AI coding session, the agent re-scans the codebase to understand it. That scan is slow, token-expensive, and inconsistent.

We pre-compute repository context once — architecture, dependencies, endpoints, schemas, cross-file reference graphs — into deterministic artifacts. Any MCP-aware host (Claude Code, Cursor, Cline, Continue, Zed) reads those artifacts instead of scanning on every prompt.

The scanner parses native ASTs in five languages via Tree-sitter grammars and ships with 290+ framework detection patterns across 14 dictionaries. A 2,780-file production monorepo scans in ~1.5 seconds. Published to npm, PyPI, crates.io, and GitHub Releases as Apache 2.0 open-source.

We built an e-commerce engine for a niche motorsport brand — van and sports car modifications — and designed every layer for reuse. The monorepo separates storefront, CMS, email templates, and UI components into packages that can be deployed under a new brand in days.

ISR product pages with generateStaticParams pre-render top products at build time across PL, EN, and DE locales, then revalidate every 2 minutes. Long-tail products render on demand, cache on first hit. Price changes in Strapi are live across all pages within 120 seconds — no deploys, no cache scripts.

Horizon Translate is embedded as the multilingual layer: product descriptions flow from Strapi in Polish and are automatically translated and cached using content-hash invalidation. Static pages are SEO-indexed in all three languages; dynamic pages translate on first render and serve from disk cache on every subsequent request.