🏛️ Making It Smooth & Trustworthy for Lawyers

🔴 TRUST BUILDERS (Immediate Impact)

1. Confidence Meter on Every Answer

• Add a confidence bar below each answer (High / Medium / Low)
• Based on reranker's top score + number of matching sources
• Color-code: 🟢 High (score > 0.7, 3+ sources), 🟡 Medium (0.4-0.7), 🔴 Low (< 0.4)
• Label: "Based on 5 documents, confidence: High (87%)"

2. Citation Inline Formatting

• Render clickable [1] superscripts inline in answer text, like legal briefs
• When clicked, open SourceViewerModal or scroll to source card
• Standard format for Westlaw, LexisNexis, Casetext

3. "Why This Answer" Transparency Box

• Collapsible box explaining how the system arrived at the answer
• Shows: chunk count, reranker score, source documents used
• Creates audit trail for court-defensible research

4. "Not in Documents" Reframed

• Change message to: "This specific question wasn't addressed in your indexed documents."
• Add: 📋 "Consider adding documents that address [topic]"
• Reframes absence as actionable intelligence

🟡 SMOOTHNESS (Reduces Friction)

5. Progressive Answer Rendering

• Semantic chunks: direct answer → supporting evidence → caveats
• Each section slides in with fadeSlideUp animation
• Mirrors how lawyers read: answer first, then authority

6. Keyboard-First Navigation

• ↑/↓ navigate source cards, Cmd/Ctrl+K command palette
• Escape close modals, Tab switch panels
• Number keys 1-5 open source citations

7. Contextual Loading States

• "🧠 Analyzing your question..."
• "🔍 Searching 18,836 document chunks..."
• "📊 Reranking top 100 candidates..."
• "✍️ Drafting response with citations..."

8. Smart Auto-Scroll

• Auto-scroll during streaming, but stop if user scrolls up
• Show "↓ New content" pill at bottom

🟢 PROFESSIONALISM

9. Document Header in Source Viewer

📄 contract_license_agreement.pdf
   Arconic Rolled Products — Trademark License Agreement
   Page 3 of 12 · Section: Territory Limitations

10. Dark/Light Mode Toggle

• Light mode: cream background, dark text (feels like paper)
• Dark mode: current design (feels like terminal)
• Persist in localStorage

11. Answer Length Controls

• Brief: 1-2 sentences
• Standard: paragraph with citations
• Detailed: full analysis with all supporting passages

12. Export as Legal Memo

MEMORANDUM
TO:       [User]
FROM:     Lawyer Assistant
DATE:     July 30, 2026
RE:       [Original Question]

ANSWER: [text]
SOURCES: [1] contract.pdf, p.5 — "The liability cap..."
CONFIDENCE: High (87%)

🔵 TRUST & COMPLIANCE

13. "Local Processing" Badge

• Persistent: 🔒 "All data processed locally — nothing sent to external servers"
• Massive differentiator vs. ChatGPT/Claude

14. Professional Disclaimer

"AI-generated responses based on your indexed documents. Always verify citations against source documents. This tool does not provide legal advice."

15. Audit Log

• History tab showing every question, sources cited, confidence level
• Creates defensible record of research process

🟣 MICRO-INTERACTIONS

16. Source Card Hover Effects

• 2-line text preview on hover
• Subtle shadow lift with translateY(-1px)
• Relevance bar animates wider

17. Staggered Answer Reveal

• 0ms: answer text fades in
• 200ms: confidence meter slides in
• 400ms: source buttons appear one by one

18. Input Bar Enhancements

• Subtle character count
• File attachment chips above input
• Rotating placeholder prompts

19. Empty State Upgrade

• Legal scales icon
• Corpus stats: "18,836 passages from 511 documents"
• Subtitle: "Every answer is cited to source text"

🏗️ Architecture Suggestions

20. Answer Pre-processing

• Check source count before LLM call
• 0 matches → show "not found" immediately
• Low confidence → prepend caveat

21. Source Deduplication

• Group multiple chunks from same file
• Show as expandable tree: 📄 contract.pdf (3 citations)

22. Skeleton Loading

• Pulsing gray lines in answer shape while LLM thinks
• Reduces perceived wait time by ~40%

Priority Matrix