Quarkus AI Text Processor with LangChain4j & Groq

Prerequisites

1. Java 17+ installed.
2. Groq API Key — obtain one at console.groq.com.

Steps

1. Clone the repository:

   git clone https://github.com/mehedicoder/langchain4j-quarkus.git
   cd langchain4j-quarkus

2. Set your Groq API key as an environment variable:

   # Linux/macOS
   export GROQ_API_KEY=your_gsk_api_key_here
   
   # Windows (PowerShell)
   $env:GROQ_API_KEY="your_gsk_api_key_here"

3. Review the tech stack: The project uses Quarkus 3.15.1, LangChain4j (Quarkus Extension), and the llama-3.3-70b-versatile model via Groq's OpenAI-compatible API.
4. Understand the architecture: AI services are declared via Java interfaces using @RegisterAiService — no boilerplate prompt handling required.
5. Build and run:

   ./mvnw quarkus:dev

   The app starts on the default Quarkus port localhost:8080.
6. Send a request to the REST endpoint:

   curl "http://localhost:8080/api/process?task=summarize&text=Your+text+here"

   Change the task query parameter to switch between summarization, translation, and other AI tasks.
7. Run tests:

   ./mvnw test

   Tests use REST-Assured and JUnit 5.

React Frontend with LangChain4j AI Backend

Prerequisites

1. Java 17+ (uses Text Blocks and Records).
2. Gradle 7+ or Maven as build tool.
3. Groq API Key set as GROQ_API_KEY environment variable.

Steps

1. Clone the repository:

   git clone https://github.com/mehedicoder/react_langchain4j.git
   cd react_langchain4j

2. Configure the Groq API key:

   # Linux/macOS
   export GROQ_API_KEY="your_gsk_key_here"
   
   # Windows (PowerShell)
   $env:GROQ_API_KEY="your_gsk_key_here"
   
   # Windows (CMD)
   set GROQ_API_KEY=your_gsk_key_here

3. Explore the three applications included:
   • ITAssistant — Terminal-based AI security consultant with streaming and chat memory.
   • ITAssistantHumanLike — Same as above with simulated human typing (Teletype effect).
   • TextSummarizer — Multithreaded text summarizer supporting 10+ summarization levels.
4. Place text files for summarization: Put .txt files in src/main/resources/ before running the TextSummarizer.
5. Build and run with Gradle:

   ./gradlew run

   Or run individual classes directly from your IDE.
6. Interact with the chatbots: Type questions at the Ask> prompt. After each response the system displays TTFT (Time to First Token) and Total Latency metrics. Type exit or quit to end.
7. Use the Text Summarizer: Run TextSummarizer.java, then follow the prompts — enter filename, summarization level (e.g., executive), and language.

Summary Level Reference

AI-Powered Invoice Verification with LangGraph4j

Prerequisites

1. Java 17+ and Maven.
2. langgraph4j dependency (included in pom.xml).

Steps

1. Clone the repository:

   git clone https://github.com/mehedicoder/invoice-verifia.git
   cd invoice-verifia

2. Review the Maven configuration: Open pom.xml to understand dependencies including langgraph4j and any LLM provider connectors.
3. Understand the architecture: The system implements a multi-agent workflow:
   • Extractor — Reads invoice data and detects amounts.
   • Router — Evaluates invoice value thresholds; low-value invoices go to the autonomous agent, high-value invoices trigger human authorization.
   • Persistence — Logs all approvals (automatic and manual).
   • Responder — Finalizes and outputs invoice status.
4. Build the project:

   ./mvnw clean install

5. Run the application:

   ./mvnw exec:java

   Observe the streaming output as invoices are processed.
6. Interact with the Human-in-the-Loop: For high-value invoices you'll see a prompt:

   Approve [A] or Reject [R]?

   Enter A to approve or R to reject. The system logs manual authorization.

Example Output

[Extractor] Detected: $45.00
[Persistence] Logged APPROVED by Agent
[Responder] Finalized Invoice: Status=APPROVED, Total=$45.00

[Extractor] Detected: $150.00
Approve [A] or Reject [R]? A
[Persistence] MANUAL APPROVAL LOGGED: $150.00 authorized by Human.

LangChain4j AI Playground — Chatbots & Text Summarizer

Prerequisites

1. Java 17+ (uses Text Blocks and Records).
2. Gradle 7+ or Maven.
3. Groq API Key set as GROQ_API_KEY.

Steps

1. Clone the repository:

   git clone https://github.com/mehedicoder/lanchain4j_playground.git
   cd lanchain4j_playground

2. Set the Groq API key:

   export GROQ_API_KEY="your_gsk_key_here"

3. Prepare text files (for summarizer): Place .txt files in src/main/resources/.
4. Run the IT Guru Chatbot:

   java org.assistant.ITAssistant

   Type security questions at Ask>. The chatbot retains 20 messages of context and streams responses. Metrics (TTFT, Total Latency) appear after each answer.
5. Run the Human-Like Assistant:

   java org.assistant.ITAssistantHumanLike

   Same as above but responses appear with simulated human typing speed via CompletableFuture.
6. Run the Text Summarizer:

   java org.assistant.TextSummarizer

   Follow the interactive prompts: enter filename, summarization level (e.g., executive), and output language. Multiple files are processed concurrently using FixedThreadPool.
7. Type exit or quit to end any chatbot session.

Technical Architecture

• Streaming: Uses LangChain4j TokenStream for reactive output.
• Concurrency: CountDownLatch and CompletableFuture coordinate async LLM responses with the synchronous CLI.
• Prompt Engineering: All prompts use Java Text Blocks. The IT Guru persona is enforced via @SystemMessage for Zero-Trust security principles.

Document Intelligence — Local RAG Knowledge Base

Prerequisites

1. Java 21+ (required for Virtual Threads / Project Loom).
2. Ollama installed and running on localhost:11434.
3. Groq API Key set as GROQ_API_KEY environment variable.

Steps

1. Install and start Ollama: Download from ollama.ai and ensure it's running on the default port 11434.
2. Pull the embedding model:

   ollama pull nomic-embed-text

3. Clone the repository:

   git clone https://github.com/mehedicoder/document-intelligence.git
   cd document-intelligence

4. Set the Groq API key:

   export GROQ_API_KEY="your_gsk_key_here"

5. Prepare your documents: Place files in a local folder. Supported formats:
   • Documents: .pdf, .docx
   • Technical: .md, .markdown, .txt
   • Data: .csv, .json
6. Build and run:

   ./mvnw clean install
   ./mvnw exec:java

   The application indexes your documents using Ollama's nomic-embed-text embeddings — all locally, no cloud uploads.
7. Ask questions: The CLI provides a "thinking" indicator followed by streaming answers. Every response includes a source citation (e.g., (Source: roadmap_2026.pdf)).
8. Ask follow-up questions: The system maintains conversation context, so you can refine queries without repeating information.

Key Features

• 100% Data Privacy: All indexing and searching happens on your hardware using local Ollama embeddings.
• Virtual Thread Performance: Java 21 Virtual Threads process massive document sets in parallel.
• Verified Citations: No hallucinations — every answer cites the source filename.
• Streaming & Thinking: Professional CLI with thinking indicators and human-like streaming responses.