Operating mode
Guided experimentation Open exploration framed by visible guardrails and source-aware review.North star
Clean, transparent, exact.
AI Garden should feel like a trusted laboratory for guided experimentation: sharp hierarchy, low-noise surfaces, and visible structure that makes model work feel deliberate instead of chaotic.
Institutional blue first Gold only for signal moments Glass over glossy Precision over spectacle
Review posture
Human-in-the-loop Outputs stay draftable, inspectable, and easy to challenge before sharing.Calm AI cue
Quiet confidence Fewer glow effects, more evidence traces, alignment, and readable system state.Palette
Official color base, tuned for a premium digital surface.
Saints Blue
#003366
Primary field color, nav, deep backgroundsCampus Blue
#1578C1
Primary interactive accent, active states, chartsGold
#F4BB33
Signal color for prompts, highlights, warning emphasisCharcoal
#444444
Dark text on light surfaces, neutral anchorsSilver
#CCCCCC
Hairlines, data dividers, frosted edge detailWhite
#FFFFFF
High-clarity text, negative space, screen lightSurface system
Frosted layers with exact edges.
Hero shell
Large clear panel with soft radial light and restrained depth.Section card
Primary content area with 1px silver line and subtle blur.Tile card
Command lanes, tools, modulesSignal chip
Gold used as a cue, not a blanketTypography
Academic clarity with a technical edge.
Display / heading
Gantari is the visual anchor.
Wide, modern, and clean enough for institutional trust while still feeling current.Data / console / preview
prompt_chain.run --mode grounded --lane faculty
IBM Plex Mono adds precision only where technical language
needs to feel exact.
Signals / labels / review states
Source pack attached Human review required Prototype only
Calm AI feels intelligent when the system surfaces what is known,
what is inferred, and what still needs a person.
Example interface moment
The command center should feel curated, not crowded.
Guided experimentation
Compare models with structure before sharing results.
The system should make the task, evidence posture, and review path visible at a glance.
Grounding required Faculty prep Balanced model
Confidence Moderate
Provenance 3 cited / 1 inferred
Share state Draft only
Current run
Evidence-first comparison 3 models x 1 prompt x curated source packSignal check
No unsupported claims detected One inference flagged for human confirmation before reuseRisk posture
Prototype only Public claims remain conservative until proxy launchModel Arena
Speed, tone, depthGrounding Lab
Sources, citations, rejection rulesWorkflow Studio
Faculty, staff, student flowsMotion
Low amplitude, high intent.
- 140-220ms for hover and selection
- Subtle rise and blur shifts over scale-heavy animation
- Panel reveals should feel measured, not playful
Imagery
Human-centered, never stock-futurist.
- Real students, faculty, labs, classroom moments
- Cool natural light with rich blue framing
- Avoid robot hands, hologram brains, generic neon cityscapes
Do not drift into
Visual anti-goals.
- Oversaturated AI gradients
- Heavy black interfaces with loud cyan glow
- Rounded toy UI that feels casual instead of precise