Design a light-mode web interface screen: clinical clarity for machine learning operations framed in cool slate, crisp card islands, and cobalt precision.
LAYOUT
The screen uses a fixed-width 240px vertical sidebar on the left and an open fluid dashboard on the right. The top bar features global search with keyboard shortcut prompts and user utility controls. Below the greeting and primary CTA, the content is orchestrated into a 4-column KPI row, followed by a 2-column split (a 65% chart/table section beside a 35% operational rail), ending with paired bottom utility modules.
PALETTE (use these exact hex values)
- #f8fafc — Canvas Background: dominant app layout background behind sidebar and card surfaces
- #ffffff — Card Surface: foreground card containers, top navigation elements, and search input fill
- #eef2f6 — Border Subtle: hairline card outlines, horizontal dividers, and progress track backgrounds
- #2563eb — Primary Cobalt: primary call-to-action button, active navigation indicator, and primary metric line
- #eff6ff — Primary Soft Tint: active sidebar nav item fill and metric icon tile backgrounds
- #10b981 — Success Emerald: positive delta text, operational status indicators, and healthy sparklines
- #f43f5e — Negative Crimson: negative trend indicators, loss trajectory chart lines, and error alerts
- #0f172a — Slate 900 Heading: page headings, primary metric numerals, and major card titles
- #64748b — Slate 500 Secondary: table column headers, timestamps, secondary navigation labels, and subtitles
- #94a3b8 — Slate 400 Muted: keyboard shortcut badges, placeholder text, and inactive chevron indicators
TYPOGRAPHY
Typeface Plus Jakarta Sans (or Inter, SF Pro Display, Manrope), weights 400/500/600/700. Sizes 11px caption, 12px body-sm, 13px body, 14px body-lg, 16px subheading, 20px heading, 26px display. Tracking -0.02em at 26px, 0 at 14px. The typographic tone is analytical, modern, and quiet, driven by a neutral neo-grotesque sans-serif with geometric legibility. Numerals are rendered in tabular figures so telemetry updates remain aligned across dense tables and KPI tiles. If Plus Jakarta Sans is unavailable, Inter or SF Pro Display provide an identical clinical and technical feel without licensing overhead.
GEOMETRY & SPACING
Corner radii — cards 16px, links 6px, inputs 10px, buttons 10px. Spacing on a 4px unit, 20px gaps, 24px between sections, 1440px container, comfortable density.
DEPTH
Depth is created almost entirely without heavy drop shadows, relying instead on high-contrast surface transitions: pure white cards set against a cool slate-gray canvas. Card borders use a 1px solid hairline (#eef2f6) with an imperceptible 0 1px 3px rgba(0, 0, 0, 0.03) shadow to prevent flat bleed. Elevated floating elements like popovers and dropdowns utilize a soft 0 10px 25px -5px rgba(15, 23, 42, 0.08) ambient blur.
IMAGERY
This interface avoids raster photographic assets entirely, utilizing precision monochrome line glyphs, colored vector sparklines, and geometric donut charts. Visual interest is derived exclusively from synthetic data artifacts—smooth Bézier curves, colored status dots, and subtle tinted icon backplates. Any photographic imagery would conflict with the clinical, tool-first AI engineering identity.
MUST
- Use solid white (#ffffff) cards against the cool slate (#f8fafc) background for all grouped content.
- Restrict bright cobalt blue (#2563eb) to the primary CTA, active navigation item, and lead chart series.
- Apply 16px border-radius consistently to all outer card containers.
- Show negative and positive deltas with matching colored micro-arrows (↑ / ↓) and 12px typography.
AVOID
- Never apply heavy black drop shadows; keep elevation borders crisp with 1px #eef2f6 and faint diffusion.
- Do not use saturated yellow, orange, or arbitrary rainbow fills across charts and tags.
- Never make chart axes or grid lines high-contrast black; use ultra-faint borders (#f1f5f9).
- Avoid sharp 0px or 4px boxy corners on cards or modal elements.
OUTPUT
Render as a clean, pixel-crisp UI design at roughly 2592:3184. Flat vector rendering, real legible text, no browser chrome, no device mockup frame, no watermark, no lorem ipsum placeholder blocks. Try in DraPub