Company Case Study

Development of a B2B SaaS Platform for AI Assistants Powered by LLM and RAG

Business Software

Case study of an enterprise AI SaaS platform designed for secure and controlled deployment of AI assistants into business processes: customer support, HR, and internal knowledge workflows built on LLM + RAG architecture.

B2B SaaS Platform for AI Assistants Powered by LLM and RAG

About the Product

AI / SaaS / Customer Support / HR Tech / Enterprise Automation • Web, API • Proprietary B2B SaaS Product

2PEOPLE ASSISTANT is a B2B SaaS platform for creating, training, and integrating AI support assistants into company business processes. The platform enables businesses to quickly launch LLM-powered AI agents that operate based on a dedicated knowledge base and plug into existing communication channels.

At the current stage, the platform fully covers the AI support assistant use case for:

Customer support
Employee support (HR)
Partner and contractor support
Product and service consultations

The product is being built and evolved as the company’s own technology platform, including analytics, architecture, UX/UI, development, deployment, and ongoing improvement.

LLM PlatformRAG SystemB2B SaaSLaunched in October 2024

Business Challenge

Companies face growing support workloads, fragmented knowledge, and difficulty scaling communication processes.

The goal of the product is to give businesses a controlled and secure way to deploy AI assistants that:

Answer strictly based on verified knowledge
Integrate into existing workflows
Scale without expanding the support team
Provide transparency, control, and analytics

Solution

We built a platform that covers the full lifecycle of an AI support assistant — from setup and training to analytics and integrations.

1. AI Assistant Creation and Configuration

Each AI assistant can be configured with:

  • Assistant tasks and role
  • Behavior rules and limitations
  • Company and product information
  • Name and communication style
  • System prompt (via UI or manual setup)

This makes it possible to build highly specialized AI agents tailored to a specific business context.

2. Knowledge Base Training and Management (RAG)

Each assistant has its own isolated knowledge base.

Supported knowledge sources include:

PDF document uploads
Website parsing
Manual content creation and editing
Wiki-style structured knowledge

Technically, the platform uses a RAG approach:

  • • content is split into chunks
  • • embeddings are generated
  • • vectors are stored in a vector database
  • • retrieved context is used to generate answers strictly within the knowledge base scope

3. Testing and Quality Control

The platform provides a test chat for every assistant, where administrators can:

Validate real user scenarios
Detect response issues and inaccuracies
Refine knowledge and settings before production launch

This reduces risk and increases business trust in the AI solution.

4. Integrations with Operational Channels

Available integrations:

Telegram Bot

The AI assistant operates directly inside a Telegram bot

Omnidesk

The assistant responds on behalf of the support team across all connected channels

Inside Omnidesk, the assistant can:

  • • respond to incoming requests
  • • automatically escalate tickets to a human agent
  • • account for triggers such as sensitive topics, missing data, and similar cases

An API is also available for custom integrations with any internal company system. The platform scales independently of any specific communication channel.

5. Analytics and Transparency

The platform gives businesses access to:

Full AI assistant conversation history
Analytics on conversations, replies, and users
Data for improving knowledge bases and assistant workflows

This turns the AI assistant from a “black box” into a manageable business tool.

6. Lead Capture (Beta)

The AI assistant can operate in the following modes:

Support operator
Service consultant

In these modes, the assistant can:

  • Recognize user intent to submit a request or lead
  • Extract and normalize data using a dedicated LLM prompt
  • Create a lead in the platform’s internal CRM

Automatic export of leads to external CRM and help desk systems is also supported.

Technology Architecture

Core Technical Solutions

RAG system for knowledge-based responses
Custom system prompts for each assistant
Dedicated LLM prompts for lead extraction and normalization
Scalable backend platform for assistant management

Technology Stack

Backend

Node.jsTypeScriptMongoDBRedis

LLM & AI

LLMQdrant (vector database)

Frontend

ReactTypeScript

Project Team

The product is being developed by an internal team:

Product Owner / Product Manager
Project Manager
Backend Developer
Frontend Developer
UI/UX Designer
QA Engineer
DevOps Engineer

Results

Production version launched

of the B2B SaaS platform

Platform is used by

real business clients

Assistants are deployed

without changing internal workflows

Support workload reduced

for support teams

Control, analytics, and scalability ensured

for AI-driven operations

Growth Roadmap

The platform is actively evolving:

Expanding the list of integrations
Adding new AI agent use cases
Enhancing analytics capabilities
Improving knowledge management tools

2PEOPLE ASSISTANT is evolving into a universal platform for the secure and controlled deployment of AI assistants into business processes.

Project Links

2PEOPLE ASSISTANT is available at:

Want to discuss your project?

Get in touch, and we’ll show you how we can help bring your idea to life