Build an AI Sales Chatbot with NLP and Product Recommendations

Create a comprehensive, production-ready AI Sales Chatbot project that meets the following objectives and technical requirements:
## Objective
Develop an AI Sales Chatbot that:
1. Understands and classifies customer queries using advanced NLP techniques.
2. Recommends products intelligently based on user input.
3. Provides natural, human-like conversational experiences.
## Technical Requirements
### 1. Data Pipeline
– Utilize the `bitext/Bitext-customer-support-llm-chatbot-training-dataset` from Hugging Face.
– Generate synthetic product catalog data or integrate real e-commerce data.
– Perform data cleaning, including:
– Removal of special characters
– Lemmatization
– Label intents such as `product_query`, `return_policy`, etc.
– Split data into training and testing sets with an 80/20 ratio.
### 2. NLP Components
– Fine-tune a `BERT-base` model for intent recognition on the Bitext dataset.
– Use spaCy to extract entities such as:
– Product types
– Price ranges
– Features
– Utilize TextBlob for sentiment analysis to detect tone.
### 3. Recommendation Engine
– Implement content-based filtering using `all-MiniLM-L6-v2` embeddings.
– Combine:
– Semantic similarity (using cosine distance)
– Rule-based filters based on price and category
– Create a hybrid recommendation approach.
### 4. User Interface
– Build the UI using Streamlit.
– Features should include:
– Chat bubbles with avatars
– A product carousel displaying images and ratings
– A sidebar with clear instructions
– Typing indicators
### 5. Code Structure
Organize the project with the following structure:
“`
ai_sales_bot/
├── data/
│ ├── raw/ # Bitext dataset
│ └── processed/ # Cleaned data + product catalog
├── nlp/
│ ├── processor.py # Text cleaning, entity extraction
│ └── intent_classifier/ # Fine-tuned BERT model
├── recommender/
│ └── engine.py # Hybrid recommendation logic
├── app/
│ ├── main.py # Streamlit UI
│ └── utils.py # Helpers for logging and formatting
├── notebooks/ # For exploratory data analysis, training, deployment
└── requirements.txt # Python dependencies
“`
## Output Expectations
### Deliverables
– A fully functional Streamlit application with a chat interface.
– A trained intent classification model achieving over 90% accuracy.
– A `README.md` containing detailed setup and usage instructions.
– Well-commented code with Google style type hints.
### Performance Metrics
– Intent classification accuracy above 90%.
– Recommendation relevance evaluated using nDCG scoring.
### Quality Standards
– Modular design separating NLP, recommendation engine, and UI components.
– Robust error handling for all user inputs.
– Scalable architecture allowing easy integration with real product APIs.
## Optional Enhancements (if time permits)
– Deploy the chatbot as a FastAPI microservice.
– Add voice input and output capabilities.
– Support multiple languages.
Use clear section headers (##) to organize your content for readability. Include specific technical details such as model names, dataset references, and evaluation metrics. Present reasoning and steps clearly before final implementation recommendations.
Would you like to focus more on UI/UX design details or deployment architecture? Please specify if so.