Skip to content

HamzaNab-Dev/chatscroll

BranchesTags

Repository files navigation

📜 ChatScroll — Every Question Becomes Lasting Knowledge

Your personal AI knowledge library — chat, save, organize, and search your knowledge forever.

🌐 Live Demo · 🏗️ AWS Architecture · 🏆 AWS H0 Hackathon 2026

Next.js ASP.NET Core Aurora PostgreSQL DynamoDB Vercel

🧪 Try It Now — Judges & Testers

Live app: https://chatscroll.vercel.app

  1. Register a free account

    • Click "Sign In" → "Sign Up" tab
    • Enter your name, email and password
    • Verify with the 6-digit code sent to your inbox
    • ⚠️ If you don't see the email, check your spam/junk folder

  2. Start chatting

    • Ask any question in the chat
    • Click "Save as Scroll" to save the best answer
    • AI will automatically suggest the right folder

  3. Explore your Library

    • Go to /library
    • Try Smart search: type "blood thinner" to find the warfarin scroll by meaning not keywords
    • Try Study Mode, Share, Write your own Scroll

  4. View AWS Architecture

All data is real — Aurora PostgreSQL + DynamoDB, no mocks or stubs in production.

🧠 What is ChatScroll?

Every day people ask AI assistants valuable questions and get great answers — then lose them forever. Chat history is linear, unsearchable, and ephemeral.

ChatScroll fixes that.

Chat with Gemini AI, save the best answers as Scrolls, and build a personal knowledge library that grows smarter over time — organized automatically, searchable semantically, and yours forever.

✨ Features

💬 AI Chat

  • Powered by Google Gemini 2.5 Flash
  • Multi-turn conversation with persistent history
  • Conversations saved to Amazon DynamoDB with 90-day TTL

📜 Save as Scroll

  • Save any AI answer with one click
  • AI automatically suggests the right folder based on topic
  • Anonymous users can save Scrolls before signing up — data migrates on registration

🔍 Smart Semantic Search

  • pgvector cosine similarity search with 3072-dimensional embeddings
  • Find "blood thinner medication" → discovers your warfarin Scroll
  • Results scoped to same folder category — no cross-topic contamination
  • Hybrid search: semantic + full-text (tsvector) combined
  • Ranking badges: ✦ #1 Match, ✦ #2 Match, ✦ #3 Match

📁 Hierarchical Folder Organization

  • AI auto-organizes Scrolls into folders like Programming → Docker
  • ltree extension for path-based folder hierarchy
  • Parent/child folder structure with toggle to show/hide subfolders

🎓 Study Mode

  • Review your Scroll library as flashcards
  • Keyboard navigation: Space to flip, ← → to navigate, Esc to exit
  • Progress bar and restart button

✍️ Write Your Own Scrolls

  • Create Scrolls from books, articles, or your own knowledge
  • Not limited to AI answers — any knowledge source works

🔗 Share & Import

  • Share any Scroll via public link
  • Anyone can import a shared Scroll to their own library

📊 Related Scrolls

  • pgvector nearest-neighbour search scoped to same folder subtree
  • Automatically surfaces semantically similar Scrolls

📤 Export

  • Browser-native PDF export (print-to-PDF)
  • Markdown export for portability

🗄️ AWS Database Architecture

ChatScroll uses two AWS databases — deliberately chosen for different data characteristics.

Amazon Aurora PostgreSQL — The Knowledge Layer

Aurora Serverless v2 with three PostgreSQL extensions working together:

Extension What it's used for
pgvector 3072-dim embeddings via gemini-embedding-001; cosine similarity search; Related Scrolls via nearest-neighbour scoped to folder subtree
ltree Folder paths as dot-separated label trees (programming.containers); subtree queries without recursive CTEs
tsvector Full-text search index; ranked with ts_rank; hybrid search combining semantic + keyword results

Aurora stores: Scrolls, folders, users (Cognito sub), conversation metadata

-- Semantic search with threshold
WHERE 1 - (embedding <=> $queryVec) > 0.5
ORDER BY embedding <=> $queryVec
LIMIT 5

-- Folder subtree query (ltree)
WHERE path ~ 'programming.*'

-- Full-text search
WHERE search_vector @@ plainto_tsquery('english', $q)
ORDER BY ts_rank(search_vector, ...) DESC

Amazon DynamoDB — The Chat Layer

DynamoDB stores the high-volume chat message stream:

Key Value
PK conversationId (String)
SK {timestamp}#{messageId} (String)
TTL expiresAt (Unix epoch, 90 days)
Capacity PAY_PER_REQUEST
  • Chronological reads and time-range queries via composite sort key
  • 90-day TTL auto-expires messages — zero infrastructure, no cron jobs
  • Scales to millions of messages with pay-per-request billing

Aurora owns structure and search. DynamoDB owns the message stream.

🏗️ System Architecture

+------------------+     +--------------------------------------+
|   Browser /      |     |           AWS Cloud                  |
|   Mobile         |     |                                      |
|                  |     |  +------------+   +--------------+   |
|   Next.js 14     |<--->|  | AWS ECS    |   |  AWS Cognito |   |
|   Vercel         |     |  | Fargate    |   |  JWT Auth    |   |
+------------------+     |  | ASP.NET 9  |   +--------------+   |
                         |  +-----+------+                      |
+------------------+     |        |                             |
| GitHub Actions   |---->|  +-----v------+   +--------------+   |
| CI/CD Pipeline   | ECR |  |   Aurora   |   |  DynamoDB    |   |
+------------------+     |  | PostgreSQL |   |  chatscroll  |   |
                         |  | pgvector   |   |  -messages   |   |
+------------------+     |  | ltree      |   |  TTL 90d     |   |
| Google Gemini    |<--->|  | tsvector   |   |  PAY_PER_    |   |
| 2.5 Flash        |     |  +------------+   |  REQUEST     |   |
| embedding-001    |     |                   +--------------+   |
+------------------+     +--------------------------------------+

CI/CD Pipeline:

git push → GitHub Actions → docker build → ECR push → ECS deploy → ✅ live

🚀 Getting Started

Prerequisites

  • Node.js 20+
  • .NET 9 SDK
  • Docker (for local backend)
  • AWS Account with Aurora PostgreSQL and DynamoDB

1. Clone the repository

git clone https://github.com/HamzaNab-Dev/chatscroll.git
cd chatscroll

2. Configure the Backend

cd backend/src/ChatScroll.Api

Create appsettings.Development.json:

{
  "ConnectionStrings": {
    "Aurora": "Host=...;Port=5432;Database=chatscroll;Username=postgres;Password=...;SSL Mode=Require;Trust Server Certificate=true"
  },
  "Gemini": {
    "ApiKey": "your-gemini-api-key"
  },
  "AWS": {
    "Region": "us-east-1"
  },
  "CognitoUserPoolId": "us-east-1_xxxxxxxx",
  "CognitoClientId": "your-client-id"
}

3. Run the Backend

cd backend
dotnet run --project src/ChatScroll.Api
# API available at http://localhost:5001

4. Configure the Frontend

cd frontend/chatscroll-web

Create .env.local:

NEXT_PUBLIC_API_URL=http://localhost:5001
NEXT_PUBLIC_COGNITO_USER_POOL_ID=us-east-1_xxxxxxxx
NEXT_PUBLIC_COGNITO_CLIENT_ID=your-client-id

5. Run the Frontend

npm install
npm run dev
# App available at http://localhost:3000

🔐 Environment Variables

Backend — appsettings.json

Key Description
ConnectionStrings:Aurora Aurora PostgreSQL connection string
Gemini:ApiKey Google Gemini API key
AWS:Region AWS region (us-east-1)
CognitoUserPoolId AWS Cognito User Pool ID
CognitoClientId AWS Cognito Client ID

Frontend — .env.local

Key Description
NEXT_PUBLIC_API_URL Backend API base URL
NEXT_PUBLIC_COGNITO_USER_POOL_ID Cognito User Pool ID
NEXT_PUBLIC_COGNITO_CLIENT_ID Cognito Client ID

🛠️ Tech Stack

Layer Technology
Frontend Next.js 14, TypeScript, Tailwind CSS
Backend ASP.NET Core 9, C#
Primary DB Amazon Aurora PostgreSQL Serverless v2
Chat DB Amazon DynamoDB
AI Chat Google Gemini 2.5 Flash
Embeddings Google gemini-embedding-001 (3072-dim)
Auth AWS Cognito + JWT
Container Registry Amazon ECR
Backend Host AWS ECS Fargate
Frontend Host Vercel
CI/CD GitHub Actions

⚠️ Disclaimer

ChatScroll is built for the AWS H0 Hackathon 2026. All AI responses are generated by Gemini AI and may contain errors. Always verify important information from authoritative sources.

📄 License

MIT License — see LICENSE for details.

Built with ❤️ for the AWS H0 Hackathon 2026 · Monetizable B2C App Track

Hamza Maher Nabelsi

🌐 Live Demo · 🏗️ AWS Architecture · 📹 Demo Video

About

AI-powered personal knowledge management app — chat with AI and save answers into an organized, searchable knowledge library. Built with Aurora PostgreSQL (pgvector), Amazon DynamoDB & Next.js on Vercel for the AWS H0 Hackathon. #H0Hackathon

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages