Remit Mind

---

🌟 Features

• Natural Language Intent Parsing: Converts casual text (e.g., "Draft a transfer of 150 USD from Alice to Bob in Mexico for family support") into immutable, type-safe Java record transaction objects.
• Relative Calendar Date Resolution: System prompts load calendar dates dynamically at request-time to resolve query relative dates (e.g., "transfer tomorrow") calendar-accurately.
• Cross-Cutting Pipeline Advisors: Implements Spring AI's Advisor pattern to cleanly separate interceptor logic:
  • PromptGuardrailAdvisor scans prompts and intercepts injection attempts before calling the LLM, protecting API budgets.
  • RequestTraceIdAdvisor injects transaction-correlation UUIDs and logs execution latency.
• Multi-Turn Conversational Memory: Uses MessageWindowChatMemory sliding windows and MessageChatMemoryAdvisor to maintain conversational context (e.g. resolving pronouns: "Actually, make that $150").
• Declarative Function Calling (Tools): Exposes Java helper services directly to Gemini as JSON Schema schemas using @Tool:
  • ExchangeRateTool fetches live exchange rates from the Frankfurter FX API.
  • CountryDataTool queries the RestCountries API to audit transfer sizes against country-specific corridor compliance limits.

---

🛠️ Architecture

The copilot decouples core AI business rules from supporting middleware interceptors and live external API lookup tools:

                  [User Request (POST /api/copilot/chat)]
                                     │
                                     ▼
                     [Advisors Pipeline Interception]
        ┌────────────────────────────┼────────────────────────────┐
        ▼                            ▼                            ▼
[SimpleLoggerAdvisor]     [PromptGuardrailAdvisor]    [RequestTraceIdAdvisor]
(Default Logging)           (Prompt Safety Check)       (UUID & Latency Timing)
        │                            │                            │
        └────────────────────────────┼────────────────────────────┘
                                     │
                                     ▼
                     [MessageChatMemoryAdvisor (RAG)]
              Reads/Writes Session Context in ChatMemory
                                     │
                                     ▼
                       [Spring AI ChatClient (Gemini)]
        ┌────────────────────────────┴────────────────────────────┐
        ▼                                                         ▼
[ExchangeRateTool]                                        [CountryDataTool]
Queries Frankfurter FX API                               Queries RestCountries API

---

🚀 Getting Started

Prerequisites

• Java 21 (or higher)
• Maven (via packaged ./mvnw wrapper)
• Google Gemini API Key configured in your environment variables:

  export GEMINI_API_KEY="your-gemini-api-key"
  

Running the Application

To build and start the Spring Boot dev server locally:

./mvnw spring-boot:run

The endpoints will be exposed at http://localhost:8080.

---

🧪 Testing

The repository contains full integration tests that exercise advisor pipelines, tool-calling networks, and conversation session persistence.

To run the test suites:

./mvnw test

Note: Live integration tests will run if a GEMINI_API_KEY is present. If the environment variable is absent, integration tests are safely bypassed while offline unit tests verify the advisors and compilation.

---

📜 License

This project is licensed under the MIT License.