🤖 FinQuery: AI Banking Assistant

FinQuery is a sophisticated Question-Answering system packaged as a Telegram bot. It leverages a fine-tuned sentence-transformer model to understand and accurately answer user questions about banking services, all built on a scalable, asynchronous architecture using Apache Kafka and Docker.

---

🌟 Key Features

• Semantic Search: Utilizes a fine-tuned bge-small-en-v1.5 model to find answers based on semantic meaning, not just keywords.
• Scalable Architecture: Employs Apache Kafka as a message broker to decouple the Telegram bot from the AI model, ensuring high performance and reliability.
• Dockerized: The entire application stack (bot, worker, Kafka) is containerized and can be launched with a single Docker Compose command.
• End-to-End MLOps: Covers the full lifecycle: fine-tuning the model, creating a search index, and deploying the inference service.

---

🏛️ Architecture

The project follows a microservice architecture:

1. Telegram Bot (telegram_bot.py): The user-facing service that receives queries from users. It acts as a Kafka Producer.
2. Apache Kafka: A message broker that buffers incoming requests.
3. Model Worker (model_worker.py): A dedicated Kafka Consumer that uses the AI model and a FAISS index to find the best answer and sends it back to the user.

---

🛠️ Setup & How to Run

Prerequisites

• Docker & Docker Compose
• Git
• A Telegram Bot Token, API ID, and API Hash.

Step 1: Clone the Repository

git clone [https://github.com/moslemamini/FinQuery.git](https://github.com/moslemamini/FinQuery.git)
cd FinQuery

Step 2: Configure Environment Variables

Create your own .env file by copying the example and then fill it with your actual credentials.

cp .env.example .env

Now, open the .env file and edit the variables (API_ID, API_HASH, BOT_TOKEN).

Step 3: Prepare Data

Place your training data (e.g., triplets.jsonl) and your knowledge base data (bank_dataset.jsonl) inside the data/ directory.

Step 4: Fine-Tune the Model (Optional)

If you want to fine-tune the base model on your own data, run the training script. This step requires a machine with a GPU.

# This command runs the training script inside a temporary Docker container
docker-compose run --rm model_worker python src/worker/train.py \
    --model_name_or_path BAAI/bge-small-en-v1.5 \
    --train_data /app/data/triplets.jsonl \
    --output_dir /app/output/finetuned-bge-model \
    --num_train_epochs 3 \
    --fp16

Step 5: Create the Search Index

This is a required one-time setup. This command will process your knowledge base and create a FAISS index for fast searching.

docker-compose run --rm model_worker python src/worker/create_index.py

This will create an index/ directory on your local machine containing the search index.

Step 6: Launch the Application

Start the entire system with one command!

docker-compose up --build

Your FinQuery bot is now live and listening for messages on Telegram. To stop the application, press Ctrl+C.

---

📜 License

This project is licensed under the MIT License. See the LICENSE file for details.