Web design company in Kolkata, drawing every screen from scratch

The single most common failure we are called in to fix is a site designed on a 27-inch studio monitor over office fibre and never once opened on the device the customer owns. In this market that device is usually an Android handset between ₹9,000 and ₹18,000, two to four years old, on a congested 4G cell, with six other apps resident in memory and the battery saver quietly throttling the CPU. Design decisions that are free on the studio monitor — a full-bleed video header, a webfont family in five weights, a parallax section, a chat widget that loads 300 KB before it draws a bubble — are not free there.

So we design to the constraint on purpose. We keep a rack of real mid-range devices and we test on them, not only in a browser emulator, because emulators model network latency but not thermal throttling or a two-year-old GPU. Layouts are designed mobile-first in the literal sense: the phone composition is drawn before the desktop one, so the phone gets the considered version rather than a squashed remainder. Tap targets are sized for a thumb on a moving bus. Contrast is checked against sunlight, not against a dark studio.

Language matters as much as layout. A large share of buyers in this city read English fluently but think in Bengali or Hindi, and they scan rather than read. That changes the design: shorter line lengths, heavier use of headings as signposts, numbers and units set in tabular figures so a specification table can be scanned in one pass, and where the audience needs it, a properly subset Bengali typeface rather than a system fallback that breaks conjuncts.

Most website design engagements in this market end with a handover of flat files, and most sites therefore start decaying the day they launch. The marketing team needs a page the comp did not cover, nobody knows what the correct heading size is, so somebody guesses. Six months later there are four button styles, three shades of the brand colour and two competing card designs, and the site looks cheaper than it did at launch even though nothing broke.

We prevent that by designing the system rather than the pages. Colour is defined as a ramp with documented roles — surface, raised surface, border, muted text, body text, brand, brand hover, focus ring — so that a new page can only be built from legal choices. Type is a fixed modular scale with defined line heights and measure limits. Spacing is a single rhythm, so vertical gaps are never invented. Every interactive component is drawn in all its states, including the ones designers routinely skip: keyboard focus, loading, empty, error, and the state where the content is three times longer than the placeholder text.

That system is then implemented as real tokens in the codebase rather than as a PDF nobody opens. When your team adds a page two years from now, the correct spacing and the correct orange are the path of least resistance, and the site stays coherent without you paying an agency to police it.

A great deal of web design quietly destroys search visibility, and the client never learns why. Headings are chosen for size rather than structure, so a page has four visual h1s and no outline. Body copy is baked into images because it looked better that way. Navigation is rendered entirely in JavaScript, so a crawler sees an empty shell. Link text reads "Read more" fourteen times, which tells a search engine nothing about any of the fourteen destinations.

Because the same people design and build here, structure is a design constraint from the first sketch. One h1 per page, and it says what the page is about in the words a buyer would use. Headings descend in order and form a genuine outline. Content is real text in the HTML, server-rendered, so it is present before any script runs. Links are written as descriptive phrases. Images carry alt text written by the person who chose the image, not auto-filled at the end.

The newer half of this is answer engines. When somebody asks ChatGPT, Gemini, Claude or Perplexity for a web design company in Kolkata, the model is working from crawled text and structured data, not from your visual design. So we mark up the things those systems need — organisation and local business data, service definitions, breadcrumb trails, FAQ blocks — and we write at least one passage per page that answers the underlying question plainly enough to be quoted. Design and discoverability stop being separate line items.

After twenty years of watching analytics on sites we designed, the pattern is boringly consistent. Redesigns that chase visual trends move conversion very little. Redesigns that remove friction and add proof move it a lot. We design for the second list.

A corporate website for an engineering firm is a credibility instrument. Its job is to survive scrutiny from a procurement committee that will read the capability pages closely, check whether your certifications are current, and forward a PDF to three colleagues. That design leans on structure, specification tables, downloadable documentation and unambiguous contact routes, and it earns nothing from animation.

An eCommerce storefront is an entirely different load. Here design is judged on catalogue navigation at scale, filter and facet behaviour, the honesty of the product page, and how few taps stand between wanting the item and having paid for it. Merchandising and page speed dominate; a 300 ms improvement in category page rendering is worth more than any hero treatment. We design storefronts with the checkout drawn first and the homepage drawn last.

A portfolio or personal brand site is a third thing again — the work is the content, so the design has to get out of its way while still establishing that a professional made it. Grid, typography and image handling carry almost the whole job, and restraint is the hardest part to sell and the most valuable to deliver.

What stays constant across all three is the method: understand the decision the visitor is making, design the shortest honest path to it, build it as a system, and measure it after launch. What changes is where the weight sits.

We run design in short, visible increments rather than disappearing for six weeks and returning with a reveal. The reveal model is theatre; it concentrates all the risk into one meeting and usually produces a round of demoralising rework because a fundamental assumption was wrong on day three and nobody caught it.

Instead you see the argument first — a structure document and wireframes that establish what each page must prove and in what order. Once that is agreed, you see a full visual direction applied to the two hardest pages, not the easiest ones, because a direction that survives the complicated pages will survive the rest. From there we build the system and roll it across the site, showing you working pages in a browser rather than images of pages.

You get a named designer and a named engineer for the whole engagement, both reachable on WhatsApp during working hours. There is no account manager relaying messages between you and the people doing the work, which is the main reason our revision rounds are short.