[P1] Add Per-Response Feedback (Thumbs Up/Down) to Chatbot

[flexykrn]

Bug Description

The AI chatbot provides answers with no mechanism for users to rate response quality. There is no way to collect granular feedback on individual responses, making it impossible to identify which answers are helpful, hallucinated, or outdated. The existing #161 mentions a feedback loop but lacks per-response granularity.

Competitor Benchmark

• ChatGPT: Thumbs up/down on every response + comment box
• Claude: 👍 / 👎 with optional detailed feedback
• Perplexity: Rate answer + flag incorrect information
• Intercom: thumbs up/down with follow-up question
• Zendesk AI: Satisfaction rating per interaction

Affected Users

• Users receiving incorrect or unhelpful answers
• Users who want to flag outdated information
• Documentation team trying to improve RAG quality

Blast Radius

• Primary: Chatbot users receiving poor answers
• Secondary: Documentation team maintaining RAG pipeline
• Impact: Data-driven improvements, reduced hallucinations, better user satisfaction

Root Cause Analysis

No feedback mechanism exists at the response level. #161 (Feedback Loop) is high-level architecture but does not specify per-response UI. Without granular feedback, the team cannot identify which documents or retrievals are problematic.

Proposed Solution

Option A: Thumbs Up/Down + Comment (Recommended)

Add to each chatbot response:

[Response text]

Was this helpful?  👍  👎
[Optional: Tell us more... _______]

Option B: Flagging System

Allow users to flag specific issues:

• ❌ Incorrect information
• ❌ Outdated content
• ❌ Missing context
• ❌ Hallucination
• ✅ Helpful

Additional Fixes Required

• Design feedback UI (subtle, non-intrusive)
• Backend storage for feedback data
• Dashboard for reviewing low-rated responses
• Integration with RAG pipeline to prioritize fixes
• Anonymous submission (no login required)

Acceptance Criteria

• Every chatbot response shows 👍 / 👎 buttons
• Optional comment field appears after rating
• Feedback is stored with response ID and timestamp
• No login required to submit feedback
• Mobile responsive placement
• Does not interfere with reading response
• Analytics dashboard accessible to docs team

Raw Context

Without per-response feedback, the team is flying blind on chatbot quality. One bad answer about staking could cost users money, but there's no way to know it's happening.