This case is my detailed account of how I designed and launched the brand website. GUAHOO (guahoo.ru) from scratch and turned it into a technology platform: with automatic pricing, 1C integrations, smart feeds for advertising systems and a computer vision pipeline that parses photos by color and model without manual routine. I focused on engineering discipline: predictable processes, transparent logs, managed schedules, flexible admins and convenient product cards that help customers make quick decisions, and below is how it works, without water and with a focus on innovation.

Starting from scratch: architecture, data model and design system

I started with designing an information architecture and product data model Thermo-steeled underwear features: rulers, seasonality, fabric composition, density, thermal insulation level, purpose (sport/city/active recreation), color variations and dimensional grids.

• Standardized attributes (color, size, fit, gender, temperature range, care, fabric %, density) – a single dictionary of values and units of measurement;
• Variative SKUs based on the basic item: size / color / stretcher - in one card, without "crippling" the range;
• Component design system: cards, spats, comparison of layers, size tables, block "how to choose a layer", iconography of fabric properties (breathing, moisture removal, elasticity, flat seam).

UI/UX was built around two goals: to maximize sensory information (info, color, landing) and to eliminate unnecessary clicks, so the product card became the center of the funnel — it is tied algorithms of price, flow, visual variation and content.

Automated pricing: rules, scenarios, schedules, logs

Inside the site, I've embedded rules and scenarios for recalculating prices, which take into account:

• Mark-ups by categories/lines, differentiated by seasonality (autumn-winter vs. spring-summer), as well as by raw materials (merinos, synthetics, mixed fabrics);
• Coefficients for color variations and sizes (rare sizes, capsule colors), as well as for sets (top + bottom);
• seasonality: RRP correction curves according to the promo calendar, "early birds" and "tail of the sale" - without manual gluing of price tags;
• Promo rules: coupons, bundles, network promotions ("-X% on the second layer", "-Y% when buying a kit"), rule priorities and protection against discount conflicts.

All counting scenarios work by schedule (Daily/Hourly/Eventual) and are accompanied by log With details: what has changed, why, with what formula and from what source (scheduled count, signal from 1C, manual trigger in the admin). In the admin, there is a “pricing tape” – a chronology of bets and rules for each SKU, so that any manager can see the story and explain the current price in 30 seconds.

Part of it is -- Operational Price Management through AdminI made clear panels: quick price increases/lowering for the attribute cluster, a simulator of what happens if you apply X coefficient to the Y line, preview feeds (below the MRC, stock intersection, zero margin corridor), all manual work is minimized: the manager sets the strategy, the system executes, monitors, logs.

Integration and migration: 1C → guahoo.ru as a reliable conveyor

I arranged. Complete migration of the catalog and attributes from 1C Key links are maintained. Important elements:

• Category mapping subcategories, taking into account the marketing navigation of the site: the category tree does not “pull” the accounting logic of 1C, but turns into a showcase structure that is understandable to the user;
• Synchronization schedulesNight incremental uploads + quick deltas throughout the day (residues/prices) so that the site always “knows the truth” about availability and value;
• Reports on the synchrons quality control: detector of missing attributes, inconsistencies of dimensional grids, empty paginations, deactivation of “broken” variations until corrected in the primary source;
• Normalization of handbooks: colors are brought to the brand dictionary (and to codes for watches), the composition of the fabric - to a single format (percentage, order), sizes - to a common standard with conversions (RU, INT, growth).

Important: Synchronization does not “break” manual marketing edits – source priorities are set up so that the manager can temporarily redefine some of the fields (e.g., SEO intro or promo text) until the next scheduled update.

“Smart” feeds and retargeting: catalog and smart formats without pain

I set it up. feeds for advertising platforms and retargeting, which are dynamically assembled from the catalog, price engine and balances, supported formats for product advertising, dynamic remarketing and smart campaigns.

• Readiness for inspections Marketplaces and advertising systems: mandatory fields (id, title, description, link, image_link, price, availability, brand), gtin/mpn where required, the correct conditionunique item_group_id for variable SKUs;
• Multifidity: different strategies for different channels (short for banners, extended for search), cutting out “gray” attributes, auto-cleaning of benefits (flat seam • thermoregulation • anti-pilling);
• Swatch exports and color variations: the ad unit leads not to a “common” card, but to a card with the selected color to reduce friction between creativity and the expected landing;
• Feeds-slices Residues and margins for “chimping” in remarketing: the system does not spend the budget on illiquid/low margins if this contradicts the strategy;
• Preview and validation inside the admin: the manager sees what exactly will go into the channel, with the illumination of problem fields.

Computer vision and AI conveyor: “Put the pictures in the folder – we will do the rest”

To speed up visual content and save managers from routine, I implemented a new approach. AI conveyor belt for photo processingThe process is that managers put the photo in a predefined folder, only the article, and then the system does everything.

What's under the hood:

• Auto-distribution by colour: computer vision model highlights the base color and its shade, compares with the brand dictionary (for example, "graphite", "steel", "sand") to associate the photo with the desired spot;
• Classification by model: detector distinguishes between “male/female/unisex”, “top/bottom/set”, “longslive/legins/thermoshtans”, “lifestyle/catalog” – and arranges the frames in the appropriate slots of the card;
• Quality controlautomatic check of sharpness, noise, exposure, repetitions; duplicates are detected through emblems (vector similarity) and do not fall into the showcase;
• Background normalization and crop under a single frame, so that the grid of cards looks sequentially;
• Generation of ALT tags and descriptions based on attributes of SKU + recognized features (for example: "Termolongslive GUAHOO, flat seam, graphite, male, merine coat 30%");
• Logistics of the conveyor: which SKU/color is attached to, which is rejected and why (resolution, broken EXIF, unrecognized color, model conflict).

This system solves two pains at once: the speed of content delivery and the uniformity of the cards, and the cards get the right color previews, galleries and confections without endless manual resaves — and the buyer sees the color / model in the window that he was interested in on the banner or in the feed.

Convenient product cards: less clicks, more context

The product card on guahoo.ru is not just a “photo and buy button”, but a decision-making center:

• Flower bundles Live gallery and URL swapping (each color is an indexable variation, without the rubble of duplicate content);
• Size grid. with tips on growth and coverage, dynamic availability of dimensions on drains (seen immediately, what is available);
• The "Why flat seam" block - briefly and in business: less friction, higher comfort, durability - with pictograms of fabric properties;
• Layer comparison (Basic/medium/external): Temperature and activity tips to assemble the task kit
• FAQ and “care for the product” with short, verifiable recommendations (washing temperature, drying, antistatic);
• Micro-CTA: "Compare", "Add to the kit", "share color" - all without unnecessary reboots.

Technically, the card uses semantic markup schema.org/Product and Offer/AggregateOffer with mandatory fields (brand, color, size, material, image, sku, gtin/mpn), which immediately affects the quality of snippets and the correctness of commodity feeds.

Performance and Stability: Core Web Vitals and Caches

Speed is part of brand trust.

• Lazy loading imaging srcset Retin; conversion to WebP/AVIF
• Bundle splitting (code-splitting), defer/async for scripts; critical content above the “bend” comes first;
• Caching at the CDN and application level, valid cache headersPreloading key fonts and images;
• Monitoring errors front and back, alarms for assynchronous falls, fallback-strategies for critical blocks of the card.

Analytics and operating panels

The site "looks at itself in the mirror" - events on the card (change of color / size, viewing the size table, opening the "Composition / Care" block, scrolling before reviews, interaction with sets) are recorded and folded into management panels. I use a single event dictionary, and I show you the decision maps in the admin: which slots are selling, where users get stuck, which sizes are more likely to lead to failures (no availability), which helps to adjust the photo rows, the checkpoints, the tips and the drains.

Admin as a management tool, not a menu item

I made the admin convenient for everyday work:

• Pricing: rules, coefficients, seasonal curves, change simulator, counting logs;
• Catalogue: attributes, cross-sella, bundles, layer tips, rapid mass edits;
• Integration: exchange schedules, queue status, log of synchrons, manual "upping" of individual SKUs;
• fida: profiles for channels, preview, validation, export journal;
• Photo conveyor: Task statuses, deviations, previews of photo → SKU/color bindings, manual adjustment if necessary.

Content and SEO as an Engineering Process

I have put together templates for categories and cards so that each page is “default correct”: readable H1/H2, unique intro, benefit blocks, FAQ, micromarkup, breadcrumbs, correct pagination and canonicals. Internal linking is built by essence: lines → categories → cards, plus “related” goods (in composition / season / purpose) – not for “tick”, but so that the buyer has a clear roadmap of choice.

Safety and compliance

HTTPS, role-based access permissions, form protection (honeypot + behavioral anti-spam filters), plug-in audits, admin logs. All background processes (exchanges, recalculations, feeds, AI tasks) are queued and have a mechanism for repeated attempts and “soft degradation” so that a failure in one place does not drop the showcase.

What ended up getting the brand

GUAHOO has not just a website, but a website. range and price management platform It's a live picture of the product that's collected and maintained automatically, and managers are no longer manual spreadsheets, they're managing strategy, and the system is executing with logs, schedules, previews, and quality signals.

Key innovations that have especially “seasoned”

1. The pricing engine with a calendar, rule priorities and a simulator – transparency and speed of reactions without manual night edits;
2. Synchronization with 1C Through clear schedules, synchronization reports, and directory normalization, data lives in the same world.
3. Smart feeds with multi-profiles and previews - compliance with the requirements of the sites and the exact hit of creatives → landing;
4. AI conveyor (colors, models, QC, ALT generation) – speed and uniformity of cards without routine;
5. Card as the center of the solution Slots, size, layers, FAQ, care, microinteractions — less clicks, more comprehensibility.

Project philosophy

I started with a simple idea: the flat-tailed thermal underwear brand does not sell "positions," but comfort and confidence So the technology platform has to do two things perfectly: give the buyer the right picture and the seller the levers to control the price, balances, content and channels. That's why there's so much focus on rules, schedules, logs and automatic quality checks. It's not the magic of IT, it's a craft: when processes are predictable and interfaces are clear.