OmniPulse AI
Request a Demo
Unified Support · Web + Telegram + AI

Enterprise Omnichannel Live Support
Where AI Meets Your Support Team

Unify website chat, Telegram personal accounts and bots in a single realtime workspace. An AI assistant powered by OpenAI API supports knowledge-grounded responses, while your team takes over complex conversations instantly.

Explore Capabilities Explore Our AI Stack →
Interactive illustration · not a live account lookup
🌐 Web Client Widget ✈️ Telegram MTProto 🤖 AI Assistant · OpenAI API
Live Channels
VIP Guest #4920 ● Online
Hỗ trợ giao dịch & tài khoản
Telegram @alex_dev 2m ago
API Documentation query
Enterprise Lead #102 5m ago
Yêu cầu demo hệ thống CRM
U
Xin chào, tôi vừa nạp 2,000 USD nhưng tài khoản chưa được cộng điểm, nhờ kiểm tra giúp tôi với!
✦
⚡ OpenAI API Assistant • RAG Context
Xin chào! Đây là tình huống minh họa. AI Assistant có thể sử dụng tri thức nội bộ để hướng dẫn kiểm tra trạng thái yêu cầu hoặc chuyển trực tiếp tới nhân viên CSKH để xác minh giao dịch trong hệ thống thật.
🤝 Test Human Handoff (Chuyển CSKH) 📝 Test AI Summary (Tóm tắt hội thoại) 🎙️ Test Voice/Media Audio Transcode
100,000+
Daily Realtime Messages
< 180ms
AI Assistant TTFT (First Token)
84%
Autonomous AI Resolution Rate
99.99%
High-Availability Uptime
Architectural Superiority

Engineered for High-Scale Enterprise Engagement

Combining OpenAI API-powered customer assistance and RAG with distributed .NET 10 services, Redis caching, MongoDB and real-time messaging infrastructure.

True Omnichannel Architecture

Single unified inbox for Web Embed Widgets, Telegram Personal accounts (MTProto worker with gap-filling & rate-limit auto-healing), and official Telegram Bots.

  • ✓ Zero missing messages with durable MTProto gap-filler
  • ✓ Seamless iOS, Android & Safari web embed support

AI Assistant & Knowledge Retrieval (RAG)

The current AI assistant uses OpenAI API with RAG-based knowledge retrieval and structured tool actions. Grounded context and review controls help reduce unsupported answers.

  • ✓ Ultra-fast streaming response via SignalR & SSE
  • ✓ Multi-lingual mastery in Vietnamese, English & Chinese

Zero-Wait Human-in-the-Loop

Seamless transition between AI and human agents. CAS distributed locks prevent race conditions, with least-busy & random tie-break distribution.

  • ✓ Instant cancellation of pending AI streams upon handoff
  • ✓ Multi-tier RBAC (SuperAdmin, Admin, Leader, Staff)

Enterprise Multimodal Media

Two-way voice message transcoding (OGG/Opus <-> MP3/WAV), progressive high-res photos, video streaming with HTTP byte-range forwarders.

  • ✓ FFmpeg background worker with moov-atom verification
  • ✓ S3 MinIO storage with secure signed asset delivery

AI-Assisted Realtime Summaries

On-demand summaries of long threads, intent extraction, sentiment analysis and shift-handover briefs using the current AI assistant integration.

  • ✓ Redis capacity gate to prevent LLM quota exhaustion
  • ✓ Supervisor QA review with structured topic tags

Modern BackOffice CMS

Full-featured management portal with live heartbeat tracking, dynamic department filtering, real-time audio waveform visualizer, and audit logs.

  • ✓ 3-layer anti-leak chat isolation architecture
  • ✓ Hot-swappable brand configurations (Multi-tenant)
Current AI Integration

AI Assistance Built Into the Support Workflow

OmniPulse currently uses OpenAI API for its AI Assistant and RAG-powered knowledge retrieval. The surrounding routing, persistence, media and human-handoff infrastructure is operated by our own services. We are evaluating Claude API as a potential future integration.

Current: OpenAI API RAG Retrieval Exploring: Claude API
  • Knowledge-Grounded Context: Retrieves relevant internal support information and conversation history for AI-assisted answers.
  • Structured Business Actions: Routes eligible requests to approved tools and internal services with authorization checks.
  • Future Model Options: Claude API is under evaluation for complex cases; no production Claude integration is claimed today.
    • AiAssistantPipeline.cs · illustrative flow ● Active Integration
    // Simplified workflow illustration (not production source) var context = await knowledgeBase.SearchAsync(message.Text); var reply = await aiAssistant.CreateReplyAsync( provider: "OpenAI API", userText: message.Text, context: context ); await foreach (var token in reply.StreamAsync()) { await _hub.Clients.Group(chatId).SendAsync("AiToken", token); } // Complex cases are escalated using the handoff service.
    Under The Hood

    Enterprise-Grade Scalable Pipeline

    High-performance distributed system engineered for low latency, zero data loss, and seamless AI-to-human workflows.

    1

    Multi-Channel Ingestion

    Web Widget (iframe), Telegram MTProto workers, and Telegram Cloud Bot ingest conversations.

    2

    SignalR WS & Routing

    Distributed .NET 10 WebSockets gateway parses intents, checks RBAC, and applies Redis locks.

    3

    AI Assistant & RAG

    The OpenAI API-based assistant uses retrieved knowledge, streams replies, and may request the handoff_to_staff action.

    4

    Durable Persistence

    Atomic CAS updates in MongoDB, MinIO media object stores, and automated summary generators.

    Workflow Comparison

    Unify AI and Human Support in One Workspace

    Compare the OmniPulse workflow with fragmented manual processes and basic rule-driven automation.

    Core Capability OmniPulse AI Fragmented support setup Basic rule-based bot
    AI Engine & Reasoning ✓ OpenAI API assistant + RAG AI available via separate integrations Fixed conversational rules
    Telegram Integration ✓ MTProto Personal Accounts + Official Bots Channel support depends on integrations Channel-dependent
    Human-AI Handoff ✓ Zero-wait CAS atomic handoff + Stream cancel Often requires separate escalation tools Escalation requires explicit setup
    Rich Multimodal Media ✓ Two-way voice transcoding, video byte-range Media flows vary by tool Often limited by bot implementation
    Vietnamese & Asian Language Nuance ✓ AI-assisted multilingual replies Translation is tool-dependent Intent handling depends on rules
    Architecture Ownership & On-Prem ✓ Full Docker/K8s deployment ready Deployment depends on vendor Varies
    Request a Product Demo

    Bring Your Support Channels Together

    See how OmniPulse combines web chat, Telegram, AI-assisted replies and seamless human takeover. Contact the LongKhang team to discuss an integration, deployment or product walkthrough.

    Request a Demo → Explore the Platform
    Knowledge Base

    Frequently Asked Questions

    Everything you need to know about our technical setup and roadmap.

    What powers the AI Assistant today?

    OmniPulse currently runs its RAG-powered AI Assistant on OpenAI API. We are exploring an optional Claude API integration for future workflows; the current website does not represent Claude as deployed.

    How does MTProto Telegram synchronization avoid flood bans?

    Our Telegram worker implements durable state tracking with exponential backoff and randomized jitter (FLOOD_WAIT 420 handler) persisted directly in MongoDB and RAM to reduce flood-limit disruptions and protect account sessions.

    How does the human handoff transition operate?

    When the AI Assistant identifies a request needing a human, it invokes the structured handoff action. Our CAS distribution algorithm assigns the ticket to the least-busy active staff member while instantly terminating active AI token streams.

    Can OmniPulse be deployed in private cloud or on-prem?

    Yes. All backend services (.NET 10 API, SignalR nodes, worker daemons, React frontends) are containerized with Docker and deployable onto private Kubernetes clusters with MinIO and Redis caching.