https://devpost.com/software/404found

404Found is powered by a network of autonomous software agents, each specializing in a part of your privacy and finance workflow. These agents are orchestrated via the Fetch.ai Agentverse platform, enabling decentralized, intelligent, and resilient operations.

User
  |
  v
[ FastAPI Backend ]
  |
  v
╔════════════════════════════════════════╗
║         Fetch.ai Multi-Agent Mesh     ║
╠════════════════════════════════════════╣
║  [ OCR Agent ]                        ║
║      |  PDF Extraction                ║
║  [ Expense Analysis Agent ]           ║
║      |  Spending Insight, Categories  ║
║  [ Gemini/LLM Insights Agent ]        ║
║      |  Advice, Drafts, Summaries     ║
║  [ Privacy Agent ]                    ║
║      |  Data Leak, Breach Checking    ║
║  [ Communication Agent ]              ║
║      |  Draft & Send Privacy Requests ║
╚════════════════════════════════════════╝
        |
        v
    MongoDB Storage

1. User Uploads Statement

   • The user submits a PDF or image of their financial statement.
   • The backend API receives the file, stores meta info, and triggers the Fetch.ai OCR Agent.

2. OCR Agent: Extracts Data

   • Uses AI to scan and extract transaction data from the PDF.
   • Sends recognized text to the next agent via Fetch.ai agent messaging.

3. Expense Analyzer Agent

   • Receives the text, parses out individual transactions.
   • Categorizes spending (e.g. food, subs, travel).
   • Detects recurring charges and potential subscriptions for cancellation.

4. LLM/Gemini Insights Agent

   • Takes categorized spend data, produces:
     • Personalized financial suggestions (saving tips, budget planning)
     • Summaries of high-/low-spending categories
     • Alerts on unusual or risky spending

5. Privacy Agent

   • Cross-references detected merchants/companies with known data brokers or breaches.
   • Flags companies known to mishandle or sell user data.

6. Communication Agent

   • Automatically drafts polite, actionable emails (via LLM) to request data deletion, subscription cancellation, or privacy policy reviews.
   • All actions are reviewed by the user before sending—empowering last-click control.

• All agents register and discover each other on the Fetch.ai Agentverse network.
• Data and requests are passed through robust, decentralized messaging.
• Agents can be swapped, retried, or upgraded independently (fault tolerance).

• At every stage, the user sees data, recommendations, and control options—nothing is sent on their behalf without a final review.
• All actions, spend patterns, and responses are stored securely in MongoDB for later review or audit.

• New agents (e.g., for investment analysis, document validation) can be plugged in instantly thanks to the modular agent protocol.
• Agents can independently use Fetch.ai’s SIone platform to access different LLMs (e.g., Qwen, Gemini) for specialized tasks.

flowchart TD
   A[User Upload] --> B[FastAPI Backend]
   B --> C[OCR Agent]
   C --> D[Expense Analyzer]
   D --> E[LLM Insights Agent]
   E --> F[Privacy/Breach Agent]
   F --> G[Communication Agent]
   G --> H[User Approval/Send]
   B -.-> I[MongoDB Storage]

• Rectangles: Agents (autonomous services)
• Arrows: Data/messages moving between agents
• Dashed Arrow: Data stored/retrieved for later use

• 404Found = Agent Mesh: Each part of the workflow is powered by an independent, upgradable agent.
• Fetch.ai = Glue: Provides registration, discovery, messaging, and LLM power for every agent.
• User-Centric: Automation where you want it, control when you need it.

For deeper details, see the backend/ai_agents directory and /docs route of the FastAPI app.

Let me know if you want any diagram as a rendered PNG/JPG or more focus on storage/integration vs. orchestration! 🚀 What’s Next for 404Found

• Plug & Play LLMs:
Opening the platform to any supported LLM (e.g., Gemini, Qwen, and more) so users can choose the best model for privacy or finance advice, or use a cheaper model for bulk tasks.

• Learning Agents:
Reinforcement learning so agents adapt to your preferences, habits, and new types of digital threats in real time.

• Agent Marketplace:
Public, user-contributed agents for things like tax optimization, investment advice, or even family/corporate privacy management—install what you want, when you want.

• Distributed Intelligence:
Federate tracking of new data leaks (e.g., real-time alerts when a broker lists your data or a new scam surfaces).

• Blockchain-powered Audit Trails:
Every privacy request and actionable outcome logged for you, and you alone, with immutable verification that can be referenced later (great for GDPR support).

• Proactive Subscription & Savings Automation:
Let agents not only suggest, but—on your terms—auto-cancel dark pattern subscriptions, set money aside towards custom goals, and notify you before you overspend.

• Seamless Frontend:
A snappier UX, including more granular notifications, progress dashboards, and privacy “health scores.”

💡 Why We Built This (Hackathon Reflection)

• Hackers for Good:
We wanted to shift AI from pure analytics or chatbots to real, actionable empowerment—giving regular people tools only insiders had before. • Modularity = Longevity:
Designing for agent modularity means others can build, fork, or plug new tech into our project. • User Control:
Automation is awesome, but only when the user can take over at any point. 404Found is designed for collaboration, not just automation.

🌍 Get Involved / Contributing

404Found is open for pull requests, issue reports, or feature suggestions! If you want to: • Build your own agent (privacy, finance, etc), see /backend/ai_agents/README.md • Fork for your org, drop us a PR or GitHub issue
• Need example REST queries? Hit the backend /docs path

⭐ Thanks & Acknowledgements

Big thanks to: • The Fetch.ai and Agentverse teams for robust, actually developer-friendly agents. • Every privacy advocate and FOSS dev who inspired this repo—this is for you!