BankSupportAI

An intelligent chatbot system for banking customer support, built with Streamlit, OpenAI Whisper, Sentence Transformers, and the Gemini LLM.
The system enhances customer interactions by providing accurate, contextually relevant responses through both text and audio outputs.

---

🚀 Key Features

• Speech Recognition – transcribe user audio inputs with OpenAI Whisper
• Intent Classification – classify queries using sentence embeddings + cosine similarity
• Contextual Response Generation – generate informed replies with Gemini LLM and chat history memory
• Text-to-Speech Output – convert chatbot responses into natural audio via gTTS

---

⚙️ Installation

1. Clone the repository

git clone https://github.com/your-username/BankSupportAI.git
cd BankSupportAI

2. Create a virtual environment (recommended)

python -m venv env
source env/bin/activate  # On Windows: env\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Set up API key Add your OpenAI API key to a .env file:

OPENAI_API_KEY="your_openai_api_key"

---

▶️ Usage

1. Run the application

streamlit run app.py

2. Interact with the chatbot

• Type your question or message in the input field
• Optionally, click the microphone icon to record an audio query

---

📂 Code Structure

• app.py – main app logic, Streamlit UI, chatbot coordination
• chatbot.py – defines the ChatBot class with LLMChain and intent classification
• audio_utils.py – handles transcription (Whisper), text-to-speech (gTTS), and audio playback
• intent.csv / intent_embeddings.csv – predefined intents and embeddings dataset

---

🔍 Technical Details

Intent Classification

• Uses sentence transformer embeddings for both queries and intents
• Cosine similarity selects the most relevant intent

Chatbot Response Generation

• ChatBot integrates with LLMChain and a custom prompt template
• Responses are context-aware, leveraging chat history + intent classification

Audio Processing

• Transcription: OpenAI Whisper converts audio → text
• Text-to-Speech: gTTS generates WAV responses, played in Streamlit

---