Designing a Trust-First Single-Screen AI Chat Interface

I designed Evidence Chat Canvas, a single-screen AI chat interface optimized for both high-trust legal Q&A and generic knowledge work. Instead of treating “AI chat” as a blank message list with a text box, I focused on the moments that actually determine whether the interface feels believable: grounding, privacy, error recovery, and motion that clarifies state changes without becoming decoration.

The primary version is framed as a Legal Q&A Assistant, not an “AI lawyer.” That decision shaped the interface: a compact trust strip surfaces jurisdiction, limitations, privacy mode, and source visibility before the user even reads the first answer. The rest of the layout stays intentionally calm — one centered reading column, a sticky file-aware composer, and inline source chips that appear exactly where trust is needed.

Visually, the system combines Material 3 structure with Apple-like restraint: an 8-point grid, soft surfaces, strong typography, semantic iconography, and subtle elevation. Motion was treated as product behavior, not ornament: sending, typing, reply reveal, chip entrances, and recoverable errors all have explicit timing and reduced-motion fallbacks. The result is a portfolio exploration that looks polished in a static frame but also feels ready for real implementation.