Feedback Pulse 📊

AI-powered product feedback aggregation and analysis tool for Product Managers

Feedback Pulse helps Product Managers consolidate scattered feedback from multiple sources, analyze sentiment and themes using AI, and make data-driven product decisions.

🎯 Overview

Product Managers receive feedback from various sources daily - customer support tickets, Discord/Slack channels, GitHub issues, social media, and app stores. Feedback Pulse solves the challenge of fragmented feedback by:

• Aggregating feedback from multiple sources into a unified view
• Analyzing each piece of feedback for sentiment, urgency, and themes using AI
• Summarizing feedback into actionable insights
• Prioritizing feature requests based on severity, quantity, and recency
• Enabling semantic search to find related feedback across all sources

✨ Key Features

Core Capabilities

• 📥 Feedback Ingestion: Unified interface for collecting feedback from multiple sources
• 🤖 AI-Powered Analysis: Automatic sentiment detection, urgency scoring, and theme extraction
• 🔍 Semantic Search: Natural language search across all feedback using vector embeddings
• 📊 Insights Dashboard: AI-generated summaries, theme distribution, and sentiment trends
• 🎯 Feature Prioritization: Smart ranking of feature requests based on multiple factors
• 💡 Idea Playground: Interactive workspace for PMs to validate feature ideas against feedback

AI-Assisted Feature Building

• Validator Mode: Validate requirements against feedback to ensure comprehensive coverage
• Sounding Board Mode: Explore creative alternatives and edge cases with AI assistance
• Conversational Refinement: Iterative chat interface for refining feature specifications
• My Features: Save and manage feature specs with linked feedback context

🏗️ Architecture

Tech Stack

Backend:

• Cloudflare Workers: Serverless API hosting and orchestration
• Hono: Lightweight, fast web framework for routing
• D1 Database: SQLite-based serverless database
• Workers AI: Built-in AI models for sentiment, summarization, and theme extraction
• Vectorize: Vector database for semantic search
• Workers KV: Key-value store for caching summaries

Frontend:

• React: UI library with TypeScript
• Vite: Fast build tool and dev server
• React Router: Client-side routing
• React Markdown: Markdown rendering for AI responses

System Architecture

┌─────────────────────────────────────────────────────────────────┐
│                      CLOUDFLARE WORKERS                         │
│                  (Hono Router + API Endpoints)                  │
│                                                                 │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐        │
│  │/feedback │  │/insights │  │ /search  │  │ /analyze │        │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘  └────┬─────┘        │
└───────┼─────────────┼─────────────┼─────────────┼───────────────┘
        │             │             │             │
        ▼             ▼             ▼             ▼
┌───────────────┐ ┌───────────┐ ┌───────────┐ ┌───────────────┐
│      D1       │ │    KV     │ │ Vectorize │ │  Workers AI   │
│  (Feedback    │ │ (Cached   │ │ (Semantic │ │ (Llama 3.1,   │
│   Storage)    │ │ Summaries)│ │  Search)  │ │  BGE Embed)   │
└───────────────┘ └───────────┘ └───────────┘ └───────────────┘

🚀 Getting Started

Prerequisites

• Node.js (v18 or later)
• pnpm package manager
• Cloudflare account (free tier works)
• Wrangler CLI

Installation

1. Clone the repository

   git clone https://github.com/prayush21/product-feedback-agent.git
   cd product-feedback-agent

2. Install dependencies

   # Install backend dependencies
   pnpm install
   
   # Install frontend dependencies
   cd frontend
   pnpm install
   cd ..

3. Initialize the database

   # Create local D1 database and run schema
   pnpm run db:init

4. Configure Cloudflare resources

   Update wrangler.toml with your Cloudflare account details:

   [[d1_databases]]
   binding = "DB"
   database_name = "feedback-db"
   database_id = "your-database-id"
   
   [[kv_namespaces]]
   binding = "CACHE"
   id = "your-kv-namespace-id"
   
   [[vectorize]]
   binding = "VECTORIZE"
   index_name = "feedback-embeddings"

Development

1. Start the backend development server

   pnpm run dev

   This starts the Cloudflare Workers dev server at http://localhost:8787

2. Start the frontend development server (in a separate terminal)

   cd frontend
   pnpm run dev

   This starts the Vite dev server at http://localhost:5173

3. Load mock data (optional)

   curl -X POST http://localhost:8787/api/feedback/bulk \
     -H "Content-Type: application/json" \
     -d @data/mock-feedback.json

📁 Project Structure

product-feedback-agent/
├── src/                      # Backend source code
│   ├── index.ts              # Main Worker entry point
│   ├── types.ts              # TypeScript type definitions
│   ├── db/
│   │   └── schema.sql        # Database schema
│   └── routes/
│       ├── feedback.ts       # Feedback CRUD endpoints
│       ├── insights.ts       # Analytics & summaries
│       ├── search.ts         # Semantic search
│       ├── analysis.ts       # AI analysis triggers
│       ├── feature-specs.ts  # Feature specifications
│       └── ai-validator.ts   # AI validation system
│
├── frontend/                 # React frontend
│   ├── src/
│   │   ├── App.tsx           # Main app component
│   │   ├── main.tsx          # Entry point
│   │   ├── api/              # API client functions
│   │   ├── components/       # Reusable UI components
│   │   ├── pages/            # Page components
│   │   ├── context/          # React context providers
│   │   ├── styles/           # Global styles
│   │   └── types/            # TypeScript types
│   └── public/               # Static assets
│
├── docs/                     # Documentation
│   ├── PRD.md                # Product Requirements Document
│   └── FEATURE_BUILDER_PRD.md # Feature Builder PRD
│
├── data/                     # Mock data for testing
│   └── mock-feedback.json
│
├── public/                   # Static assets served by Workers
│   └── index.html
│
├── package.json              # Backend dependencies
├── wrangler.toml             # Cloudflare Workers config
└── tsconfig.json             # TypeScript config

🔌 API Endpoints

Feedback Management

Method	Endpoint	Description
GET	/api/feedback	List feedback with filters (source, sentiment, date)
GET	/api/feedback/:id	Get single feedback item with analysis
POST	/api/feedback	Create new feedback item
POST	/api/feedback/bulk	Bulk import feedback
DELETE	/api/feedback/:id	Delete feedback item

Analysis

Method	Endpoint	Description
POST	/api/analyze/:id	Trigger AI analysis for specific feedback
POST	/api/analyze/batch	Analyze all unprocessed feedback

Insights & Summaries

Method	Endpoint	Description
GET	/api/insights/summary	AI-generated summary (daily/weekly/custom)
GET	/api/insights/themes	Top themes with counts and sentiment
GET	/api/insights/priorities	Prioritized feature requests
GET	/api/insights/sentiment-trend	Sentiment over time
GET	/api/insights/sources	Breakdown by feedback source

Semantic Search

Method	Endpoint	Description
GET	/api/search?q=...	Natural language search across feedback
POST	/api/search/similar/:id	Find similar feedback items

Feature Specifications

Method	Endpoint	Description
GET	/api/feature-specs	List all feature specifications
GET	/api/feature-specs/:id	Get single feature spec
POST	/api/feature-specs	Create new feature spec
PUT	/api/feature-specs/:id	Update feature spec
DELETE	/api/feature-specs/:id	Delete feature spec

AI Validation

Method	Endpoint	Description
POST	/api/ai/validate-feature	Validate requirements against feedback

🗄️ Database Schema

The application uses a relational schema with the following tables:

• sources: Feedback sources (Discord, GitHub, Twitter, etc.)
• feedback: Main feedback items with AI analysis results
• themes: Extracted themes/topics
• feedback_themes: Many-to-many relationship between feedback and themes
• feature_requests: Extracted feature requests with priority scores
• feature_specs: PM-created feature specifications with conversation history

For the complete schema, see src/db/schema.sql.

🤖 AI Models

Task	Model	Purpose
Sentiment Analysis	@cf/huggingface/distilbert-sst-2-int8	Classify sentiment as positive/negative/neutral
Summarization	@cf/meta/llama-3.1-8b-instruct	Generate one-line summaries
Theme Extraction	@cf/meta/llama-3.1-8b-instruct	Extract key themes from feedback
Urgency Detection	@cf/meta/llama-3.1-8b-instruct	Classify urgency (critical/high/medium/low)
Embeddings	@cf/baai/bge-base-en-v1.5	Generate 768-dim vectors for semantic search
Feature Validation	@cf/meta/llama-3.1-8b-instruct	Validate requirements & provide suggestions

📊 Deliverables

1. Dashboard

• Real-time metrics: total feedback, average sentiment, top themes
• AI-generated daily/weekly summaries
• Recent feedback with sentiment badges
• Theme distribution visualization

2. Feedback List & Detail Views

• Filterable feedback list (by source, sentiment, urgency, date, theme)
• Detailed feedback analysis with AI insights
• Similar feedback suggestions using semantic search

3. Insights & Analytics

• AI-generated insights with key takeaways
• Theme breakdown with drill-down capability
• Feature request prioritization
• Sentiment trends over time

4. Idea Playground (Feature Builder)

• Interactive sidebar for exploring feedback
• AI-assisted feature validation in two modes:
  • Validator: Ensure requirements comprehensively address feedback
  • Sounding Board: Explore creative alternatives and edge cases
• Conversational refinement with chat interface
• Save feature specs to "My Features" library

5. Semantic Search

• Natural language query support
• Vector-based similarity matching
• Context-aware results

🚢 Deployment

Deploy to Cloudflare Workers

1. Deploy the backend

   pnpm run deploy

2. Initialize remote database

   pnpm run db:init:remote

3. Build and deploy frontend

   cd frontend
   pnpm run build
   # Copy build output to ../public/
   cp -r dist/* ../public/
   cd ..
   pnpm run deploy

Environment Configuration

Set up the following in your Cloudflare dashboard:

1. D1 Database: Create a D1 database and update database_id in wrangler.toml
2. KV Namespace: Create a KV namespace for caching and update id in wrangler.toml
3. Vectorize Index: Create a Vectorize index named feedback-embeddings
4. Workers AI: Enabled automatically with Workers

🧪 Testing

# Type checking
pnpm run typecheck

# Frontend tests
cd frontend
pnpm run build
pnpm run preview

📝 Usage Examples

Load Mock Feedback

curl -X POST http://localhost:8787/api/feedback/bulk \
  -H "Content-Type: application/json" \
  -d @data/mock-feedback.json

Analyze All Feedback

curl -X POST http://localhost:8787/api/analyze/batch

Search Feedback

curl "http://localhost:8787/api/search?q=audio+quality+issues"

Get Daily Summary

curl "http://localhost:8787/api/insights/summary?period=daily"

🛠️ Development Workflow

1. Make changes to backend code in src/ or frontend code in frontend/src/
2. Hot reload is enabled in both dev servers
3. Test locally using the development servers
4. Type check with pnpm run typecheck
5. Deploy when ready using pnpm run deploy

📚 Documentation

• Product Requirements Document - Complete product specification
• Feature Builder PRD - Idea Playground feature spec
• Cloudflare Workers Docs
• Hono Documentation

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is part of a demonstration/educational repository.

👥 Authors

• @prayush21

🙏 Acknowledgments

• Built with Cloudflare's Developer Platform
• Powered by Llama 3.1 and BGE embeddings
• UI inspired by modern product management tools

---

Demo Context: This application demonstrates AI-powered feedback analysis for NotebookLM, an AI-powered notebook/research assistant. The principles apply to any SaaS product requiring centralized feedback management.