This post was drafted with the help of AI from our own commit history and reviewed by a person before publication; illustrations are AI-generated. It may contain errors and is not professional advice. Report a problem: info@delionsoft.com.
A dense week: the new company site went to production, the blog got a new structure, Imperecta’s parser core moved to Rust and its alert engine went live, and a new product entered the organisation. Here is what we decided and why.
delionsoft.com
The redesign is live: a new layout with light and dark themes, three languages (English, Ukrainian, Romanian), product and trust pages, and a full SEO pass — per-page metadata, structured data, sitemaps and hreflang for every locale.
Performance was a project of its own. Properly sized icons instead of a heavy logo file, fewer font preloads, animations moved to the compositor: mobile Lighthouse went from the high forties to the eighties, desktop sits at 99. Social cards now derive their title and description from each page, so a link to /trust or /products previews as that page rather than the home page.
The blog became Orbit, with three rubrics: Radar for industry news worth a technical lead’s minute, this Build log, and Deep dives. Every Radar post carries its source, and Ion — Orbit’s editor — appears with an expression chosen per story. Illustrations are art-directed per post and rotate through two dozen visual forms, so the feed does not converge on one look; image URLs are versioned by content, so a replaced illustration shows immediately.
Imperecta
The parser core is now Rust. We moved the computational core of our data collector — price and currency parsing, structured product-data extraction — from Python to Rust while keeping the results bit-for-bit compatible (74 parity tests over a shared corpus). The hot operations got several times faster: price parsing 21×, currency detection 15×, extracting a product card from markup 5×. Alongside the new core, a Rust data-quality validator went live: every collected record receives a 0–100 score and a set of flags before it reaches the database, so only complete and consistent data enters the pool. For customers this means fresher prices on the same collection schedule and noticeably cleaner datasets — no phantom prices, mixed-up currencies or empty product cards.
Alerts are live. A rule engine evaluates price movements and availability transitions with cooldowns, writes an append-only event stream and delivers notifications. A separate operational channel covers the collection, parsing, quality and market-data modules with in-process rate limiting, and the admin interface gained an operations overview with live counts per module and a resolve action that goes through the same guarded write path as every other change.
Sturdier foundations. Serialization is now driven by column types rather than by the shape of the value, which closes a whole class of JSON-versus-array mismatches, and rejected payloads are recorded reliably. The backend test suite runs on every push: over a thousand unit tests, with the hundred-odd that need real services separated into an integration stage.
FYM: Fit Your Mind
A new product joined the organisation this week and received its first README — the document that fixes what it is for, how it is built and where it stands. We are deliberately not describing it in detail yet. In one sentence: FYM — Fit Your Mind — is a mobile companion for mental and cognitive fitness, designed privacy-first, with the same guardrails-before-features discipline we apply everywhere else. More once there is something to show.