Introduction

AloSallam is an AI-powered sales assistant designed for sellers on Basalam. The platform helps sellers automatically respond to customer questions by combining product information, frequently asked questions, previous seller responses, and AI-powered retrieval.

The system is designed to reduce the time sellers spend answering repetitive questions while keeping responses relevant to the seller's actual products and knowledge.

Features

• AI-powered customer support for Basalam sellers.
• Automatic customer question answering based on retrieved knowledge.
• RAG architecture for retrieving relevant product and seller information before generating an answer.
• Product knowledge indexing to make seller products available to the AI.
• Seller answer learning, allowing manually answered questions to become reusable knowledge.
• FAQ and knowledge-base retrieval for frequently asked customer questions.
• Confidence Score for generated answers to determine whether an answer is reliable enough to send automatically.
• Seller referral system for questions that require human intervention.
• SMS notifications when an important question cannot be confidently answered by the AI.
• Multi-tenant architecture with isolated data and settings for each seller.
• Seller management panel with statistics and system activity.
• JWT-based authentication for secure access.
• Responsive and mobile-friendly interface designed for sellers who primarily use mobile devices.

AI & RAG Architecture

One of the core parts of AloSallam is its Retrieval-Augmented Generation architecture.

Before generating an answer, the system searches the seller's knowledge base for relevant information. This knowledge can include:

• Product information.
• Previous seller answers.
• Frequently asked questions.
• Other seller-specific knowledge.

If a relevant source cannot be retrieved, the system does not blindly ask the language model to generate an answer. Instead, the question can be referred to the seller.

For generated responses, the system also tracks information such as confidence, retrieved sources, generation time, and answer metadata.

This approach helps reduce unsupported AI responses and keeps generated answers grounded in the seller's actual data.

Technical Details

• Frontend: Next.js, React, TypeScript
• Styling: Tailwind CSS
• Backend: Node.js / NestJS
• Database: MongoDB
• AI: OpenAI API
• Architecture: Retrieval-Augmented Generation (RAG)
• Authentication: JWT
• Notifications: SMS
• Data Isolation: Multi-tenant architecture
• Deployment: Linux server with Docker

Challenges

One of the main challenges was building an AI system that could answer customer questions accurately without relying only on the language model's general knowledge.

The solution required designing a retrieval pipeline that first searches seller-specific information and only generates an answer when sufficient relevant context is available.

Another challenge was designing the system as a multi-tenant platform where each seller has independent products, knowledge, settings, conversations, and AI responses.

The platform also needed to handle uncertain questions safely. Instead of generating potentially incorrect answers, the system uses retrieval results and confidence information to determine when a question should be passed to the seller.

Outcome

AloSallam combines modern full-stack development with AI and RAG technologies to create an automated sales assistant for Basalam sellers.

The project demonstrates experience in building AI-powered applications, retrieval pipelines, multi-tenant architectures, API-driven systems, authentication, database design, and production deployment with Docker.