Back-end

Architecture & Logic

Backend Overview

The backend is designed as a lightweight, modular canister running on the Internet Computer. Its main purpose is to combine AI-powered conversations with simple account management, all wrapped in a clean and extendable architecture.

Core Ideas

• AI Interaction Layer Powered by the ic-llm library, the backend can handle both single prompts and multi-message chats using the Llama 3.1 model. This makes it easy to build interactive, context-aware experiences without external AI servers.

• Account Management Layer A simple but functional in-memory system keeps track of unique account addresses. This demonstrates how user-related data can be stored and retrieved efficiently on the IC.

Architecture at a Glance

1. lib.rs — The central entry point. Defines the public methods the frontend can call, such as sending AI prompts, managing conversations, and handling account operations.

2. account.rs — A focused module for account creation and retrieval. Uses a lightweight in-memory state to keep the logic clear and maintainable.

3. backend.did — The Candid interface. Describes the data structures and public methods so the frontend and other canisters know exactly how to communicate with the backend.

4. Cargo.toml — The project manifest. Lists dependencies (ic-cdk, ic-llm, candid) and sets up the backend to compile as a canister.

How It Works in Practice

• The frontend connects to the backend via Candid.

• When a user interacts:

  • For AI: The backend sends their prompt or conversation history to the LLM and returns the AI’s reply.

  • For accounts: The backend adds or retrieves addresses in its state.

• All responses are immediate, with no need for an external database in this version.