AI-driven CRM and Sales Marketing: Chatbot and Recommendations

🌐 This internship project, undertaken as part of a university program and approved by both academic and company supervisors, addressed the challenges faced by Small and Medium Enterprises (SMEs) in Malaysia in leveraging AI for Customer Relationship Management (CRM) and sales marketing within Enterprise Resource Planning (ERP) systems.

🌐 Developed as a Proof of Concept (POC) for the company’s own Point of Sales (POS) platform, the project aimed to explore the integration of AI-driven solutions to support business clients in automating operations, enhancing customer engagement, and improving operational efficiency. Functional APIs and a GenAI chatbot were successfully developed and validated based on the company’s existing database.

The project is structured into 5 key AI-driven components:

🌐 Gemini GenAI Chatbot by RAG
Developed an AI Chat Assistant using the Google Gemini LLM API and Retrieval-Augmented Generation (RAG) approach. It processed unstructured data (FAQs, user manuals) by feeding relevant materials to Gemini to provide accurate and grounded responses. It also offers optional support channels like phone call, email, WhatsApp, and YouTube video tutorials for further assistance.

🌐 Customer Review Sentiment Analysis
This involves extracting customer review data from database. A pre-trained BERT model classified reviews into five sentiment categories, while K-means clustering and LDA topic modeling were applied to identify key themes within semantically similar review groups, providing a nuanced understanding of customer feedback with emotional context.

🌐 Product Recommendation System
Designed a product recommendation system with two different approaches. The Content-Based Filtering (CBF) system uses cosine similarity to generate personalized recommendations. The Collaborative Filtering (CF) system leverages Truncated SVD on a user-item matrix to identify hidden preferences and predict user-item scores, leading to tailored product suggestions.

🌐 Bundle Marketing Sales Analysis
Developed a bundle sales marketing system using association rule mining. This involved extracting historical transaction data from the database. The Apriori algorithm was applied to identify frequent itemsets and calculate support, confidence, and lift metrics, revealing valuable product combinations.

🌐 Customer Spending Power Prediction
Trained a customer spending power prediction model using historical customer data. A XGBoost classifier was trained to categorize customer spending into four distinct price groups. This prediction capability enables targeted marketing efforts based on different customer spending behaviors.

💬 Tech Stack:
- Programming Languages: Python, JavaScript, HTML & CSS
- Web Framework: FastAPI, Streamlit
- Database: Microsoft SQL Server
- Machine Learning Algorithms: K-means Clustering, Cosine Similarity, Truncated SVD, Apriori, XGBoost
- GenAI: LangChain, ChromaDB, Google AI Studio, Google Gemini (LLM), GCP
- NLP: Hugging Face Transformers (BERT, MiniLM), Sentiment Analysis, LDA Topic Modelling
- Approaches: ETL, Retrieval-Augmented Generation (RAG), Content-Based Filtering (CBF), Collaborative Filtering (CF), Association Rule Mining

*Note*: Only system architecture diagrams are shown due to company privacy.