Project Description
AST AI Agents – CS RAG Agent is an AI-powered customer support solution designed to help businesses automatically answer customer inquiries using their existing knowledge and support data.
The system uses Retrieval-Augmented Generation (RAG) to retrieve relevant information from FAQs, user manuals, and historical customer support data before generating responses. By grounding answers in a dedicated knowledge base, the chatbot can provide more relevant and context-aware support while reducing reliance on general AI-generated knowledge.
Relipa was responsible for the end-to-end development of the solution, including the RAG pipeline, chat interface, response evaluation mechanism, and Operator handoff flow, built on Amazon Bedrock Knowledge Bases with a flexible integration architecture.
Project Information
- Client Name: Confidential
- Service: AI
- Platform: RAG-based AI Customer Support System
- Year: 2026
Results and Benefits
CS RAG Agent enables businesses to introduce AI-powered customer support while continuing to use their existing knowledge, websites, and operational systems. By retrieving information from predefined business data before generating answers, the system improves the consistency and reliability of customer support responses, particularly for questions that need to be answered based on existing documentation or support records.
The solution also gives businesses greater control over the data used by the AI system, with data management, customer data isolation, personal information protection, and access control incorporated according to the implementation environment. For daily operations, automating frequently asked questions reduces the repetitive workload on support teams, allowing human staff to focus on cases that require further investigation or judgment – while the flexible integration architecture lets businesses adopt AI Customer Support without replacing their existing systems.
Client Request
The client wanted to develop an AI chatbot capable of automatically receiving and responding to customer support inquiries.
Instead of relying only on a general-purpose Large Language Model, the chatbot needed to generate responses based on the client’s existing customer support information, including FAQs, user manuals, and historical support data. These information sources needed to be organized within a RAG Knowledge Base so that the system could retrieve relevant information before generating a response.
The key requirements included:
- Automatically answering customer support questions based on FAQs, User Manuals, and historical customer support data
- Bringing these data sources into a RAG Knowledge Base to support information retrieval and response generation
- Building a chat interface that allows users to receive and communicate with the system using natural language
- Transferring the request to an Operator when the chatbot is unable to answer
- Supporting customization and integration with various systems as needed
The client also needed the solution to remain flexible enough to adapt to different environments rather than being limited to a standalone chatbot application.
Development Process
STEP 1
Requirement Analysis
Relipa worked closely with the client to identify expected use cases, the customer support workflow, available data sources, and the functional scope of the CS RAG Agent, including the conditions under which the AI should answer automatically or transfer a conversation to an Operator.
STEP 2
Solution Design
Based on the requirements, Relipa designed the overall system architecture - how business data would be stored and retrieved through the Knowledge Base, how the AI model would generate responses, and how the system would connect with the chat interface and existing business environments. The proposed architecture was reviewed and confirmed with the client before implementation.
STEP 3
Implementation
Relipa developed the CS RAG Agent end to end, including the RAG pipeline, chat interface, system integration components, conversation context management, response evaluation mechanisms, and the Operator handoff flow.
STEP 4
Testing and Feedback
Testing was conducted continuously throughout development to evaluate information retrieval, generated responses, conversation behavior, and system integration. Results were consolidated and reported to the client, and the team continued adjusting retrieval behavior, response generation, and system configuration based on the findings and client feedback to improve the overall performance and reliability of the solution.
Tech Stacks we use
TypeScript
Python
SQL
NestJS
TYPEORM
React
Tailwind CSS
Recharts
React Markdown
AWS Bedrock
LangChain
PostgreSQL
Pandas
Redis
Jest
Vitest
JavaScript
AWS SDK v3
Next.js 14
Ragas 0.4+
Solutions
[1]
RAG Knowledge Base for Business-Specific Customer Support
One of the key challenges was enabling the chatbot to answer questions based on the client’s actual business information instead of relying only on the general knowledge of a Large Language Model. To solve this, Relipa built a RAG-based architecture that connects FAQs, user manuals, and historical customer support data to a dedicated Knowledge Base. When a user submits a question, the system retrieves relevant information from the Knowledge Base and uses it as context to generate a response – allowing the chatbot to provide answers closely aligned with the company’s actual information rather than generic AI knowledge.
[2]
Amazon Bedrock Knowledge Bases Integration
To support this retrieval workflow, CS RAG Agent uses Amazon Bedrock Knowledge Bases as the backbone of its information retrieval and AI response pipeline. By separating the knowledge source from the conversational interface, businesses can continue managing and updating their information independently while the AI agent draws on that knowledge during customer interactions.
[3]
Response Reliability Evaluation
Generating an answer is not always enough for a customer support system – the system also needs to judge whether the available information is sufficient to support that answer. To address this, Relipa built in a response reliability evaluation mechanism that checks retrieved information before delivering a result. When sufficient supporting information cannot be identified, the system avoids giving an unsupported answer and instead moves the conversation to the appropriate fallback workflow.
[4]
AI-to-Human Operator Handoff
Not every customer inquiry can or should be handled entirely by AI. When CS RAG Agent cannot identify sufficient information to respond appropriately, or when human assistance is required, Relipa built a handoff mechanism that transfers the conversation to an Operator – creating a support workflow where AI handles routine information requests while human staff remain available for more complex or exceptional cases.
[5]
Flexible Integration with Existing Systems
A further requirement was that the chatbot operate within the client’s existing digital environment rather than requiring a completely new platform. Relipa designed CS RAG Agent so it can be integrated into websites or other business systems through an appropriate interface and integration mechanism, with the chat widget and surrounding experience customizable to the business’s website, brand guidelines, and operational requirements – minimizing unnecessary changes to current systems.
[6]
Data Management and Access Control
Because customer support systems handle business documents, support histories, and other sensitive information, Relipa built in mechanisms for managing data access, separating customer data, protecting personal information, and controlling how different users or systems interact with available information – configurable according to the requirements of each implementation environment.
Gallery
Product Features
RAG-based Customer Support
The system retrieves relevant information from FAQs, user manuals, and historical customer support data before generating answers to customer inquiries, helping provide more relevant and consistent support.
Direct Integration with Websites and Existing Systems
CS RAG Agent can be integrated directly into a company's website or existing system, allowing businesses to introduce AI Customer Support without replacing their current platform.
Customizable Chat Widget
A customizable Chat Widget can be adapted to the design and branding of the company's website, keeping the AI experience consistent with the existing digital interface.
Response Confidence Evaluation
The system evaluates available information before generating the final response, helping limit situations where the chatbot answers without sufficient supporting information.
Multi-method Information Retrieval
CS RAG Agent combines different retrieval approaches to locate relevant context from the Knowledge Base for different types of customer questions.
Conversation Context Management
The chatbot maintains information from previous messages within a conversation, allowing it to interpret follow-up questions based on existing context.
Operator Handoff
When the chatbot cannot appropriately handle an inquiry or human assistance is needed, the conversation can be transferred to an Operator, letting AI and support teams work within the same process.
A Partner Invested in Your Long-Term Success
