Real-time conversational AI platform
A production conversational-AI backend that streams answers in real time, remembers the conversation, cites every source, and works across any LLM provider.
The problem
A huge consumer audience wanted instant, trustworthy answers to open-ended questions, with live information, cited sources, and multiple languages, at a speed and scale that static pages and off-the-shelf chatbots couldn’t touch.
What we built
We built a production conversational-AI backend that streams answers in real time, remembers the conversation, and works across any LLM provider, powered by LangChain and LangGraph agents on an async FastAPI service. Behind it, a distributed engine searches, crawls, and retrieves live sources over the Model Context Protocol, reranks what matters, and cites every source automatically. The same multi-tenant foundation now serves more clients, including on WhatsApp.
Results
- Real-time streaming chat with automatic source citations and multilingual, follow-up-aware answers
- Agentic tool use (live web search, crawling, and RAG retrieval) orchestrated with LangGraph
- Horizontally scaled, containerized services with queue-based distributed processing built for production load
- One multi-tenant codebase reused across multiple clients and channels, including WhatsApp
Deployment
Docker Compose deployment with horizontally scaled workers, rate limiting, and hardened security headers, deployable inside your own cloud.