| Section | Description |
|---|---|
| ⚡ Overview | Project introduction and badges |
| 🌪️ Problem | The email overload challenge |
| 💡 Solution | How SmartInbox AI solves it |
| 🛠️ Tech Stack | Technologies used |
| 🎮 Demo | Live workflow visualization |
| 🧠 Logic Flow | How the system works |
| 🚀 Quick Start | Setup instructions |
| 💡 Why This Project | Key demonstrations |
| 🔧 Use Cases | Practical applications |
| 📊 Workflow Details | Node breakdown |
| 🌐 Connect | Social links |
We receive 120+ emails daily. 80% are noise. Traditional filters are "dumb"—they either miss sophisticated phishing or accidentally archive important client communications.
An autonomous Human-in-the-Loop (HITL) orchestration that uses LLMs to think like a personal assistant. It doesn't just filter; it reasons through content to determine intent.
- Orchestration: LangFlow / Flowise (LLM Workflow Builder)
- Brain: OpenAI GPT-4o Model
- Integrations: Gmail API (OAuth 2.0)
- Logic: Custom JSON Output Parsing & Conditional Routing
.gif)
Real-time execution of the AI Agent analyzing and routing incoming traffic.
The flow triggers on every new Gmail thread. The Spam Analyzer Agent extracts the intent, urgency, and sender credibility.
Using a Routing Node, the system forks into three distinct outcomes based on the AI's classification:
| Outcome | Action taken | Security Level |
|---|---|---|
| 🚨 Spam | Prepares deletion message & waits for human approval. | High |
| ✅ Legitimate | Marks as Read + Triggers an AI-generated draft reply. | Low |
| Keeps in Inbox for manual review (Safety Net). | Critical |
Unlike basic scripts, this project features a Wait for Approval node. This demonstrates an understanding of Enterprise AI Safety—ensuring the AI never performs a destructive action (deletion) without a final human confirmation.
- Clone the Repository:
git clone https://github.com/manas-shukla-101/SmartInbox-AI.git
- Import JSON Configuration:
- Open your dashboard (LangFlow/Flowise).
- Click Upload/Import and select the SmartInbox-AI file.
- Connect API Credentials:
- Input your OPENAI_API_KEY or change it to gemini model and use Gemini API.
- Authenticate the Gmail Trigger node via Google Cloud Console.
- Deploy:
- Hit Publish. Your agent is now live.
This project demonstrates:
- AI Integration: Real-world OpenAI implementation
- Workflow Design: Complex logic with approval gates
- Error Prevention: Human-in-the-loop for critical actions
- Scalability: Modular nodes for easy extension
- Best Practices: Clean separation of concerns
- Personal email management
- Business spam filtering
- Automated customer support
- Email workflow automation
- AI prototype testing
| Node | Function | Type |
|---|---|---|
| Gmail Trigger | Listens for new emails | Trigger |
| SpamAnalyzer | Classifies content | AI Agent |
| Route by Classification | Directs flow | Router |
| Manual Approval | Human verification | Approval |
| Delete Spam | Removes spam | Action |
| Mark as Read | Processes legitimate | Action |
| Send Reply | Auto-responds | Action |
Made with ❤️ by Manas Shukla
