ChatSphere

Overview

ChatSphere is a production-grade AI chat assistant platform delivering ChatGPT-like experiences with multi-model support, session-based memory, file intelligence, and streaming responses — engineered for enterprise deployment.

The Challenge

Businesses needed an AI assistant that could handle multiple LLM providers, maintain conversation context across sessions, process uploaded documents (PDF, DOCX, Excel, images), and deliver responses with the fluidity users expect from modern AI tools — all within a secure, scalable backend architecture.

Our Solution

We built ChatSphere on ASP.NET Core with Microsoft Semantic Kernel as the AI orchestration layer. The platform routes requests intelligently across OpenAI, Azure OpenAI, and Google Gemini based on task requirements. Session memory powered by IMemoryCache with Redis-ready distributed caching ensures conversation continuity. File intelligence pipelines parse documents and images before feeding context to the AI models.

Architecture

User Interface (Razor Chat UI) → ChattingController → Session Cache (IMemoryCache / Redis) → AI Router (OpenAI / Gemini / Azure) → Semantic Kernel ChatCompletionService → Streaming Response Pipeline → Markdown + Syntax Highlight Rendering (Marked.js + Highlight.js)

Key Features

• Real-time streaming AI responses (token-by-token output)
• Multi-model AI routing (OpenAI, Azure OpenAI, Gemini)
• Session-based conversation memory
• File intelligence: PDF, DOCX, Excel, images, and code files
• Markdown rendering with syntax-highlighted code blocks
• Configurable system prompts and prompt engineering
• Secure backend with Identity-ready authentication
• Google Sign-In with OAuth 2.0
• Passkey authentication support
• Distributed caching architecture (Redis-ready)

Business Impact

• Reduced time-to-answer for internal knowledge queries by 80%
• Enabled document-based AI interactions without manual data entry
• Provided a reusable AI platform architecture for multiple client deployments
• Demonstrated production viability of multi-model AI routing

Results

• Sub-second first-token response times with streaming architecture
• Successful processing of PDF, Word, Excel, and image file uploads
• Seamless model switching without user-visible disruption
• Secure authentication with Google OAuth and passkey support

Future Scope

• RAG integration with enterprise knowledge bases
• Admin dashboard for usage analytics and model cost management
• Plugin marketplace for domain-specific AI capabilities
• Multi-tenant SaaS deployment model