GitHamza0206/simba
OpenSource Production ready Customer service with built in Evals and monitoring
What it solves
Simba provides a high-efficiency, open-source customer service assistant that eliminates the "black-box" nature of many AI solutions. It solves the problem of generic AI responses and vendor lock-in by allowing teams to have full control over their data, the RAG pipeline, and the ability to measure AI quality through built-in evaluation metrics.
How it works
Simba uses a Retrieval-Augmented Generation (RAG) architecture. It consists of a FastAPI backend that connects a website (via an npm chat widget) to a vector store (such as Qdrant or FAISS) and an LLM (OpenAI or local models). It employs Celery and Redis for asynchronous document ingestion. Users can customize every part of the pipeline, including embedding models, chunking strategies, and rerankers.
Who it’s for
It is designed for teams who want to deploy a self-hosted, customizable AI customer service assistant on their website and need to measure performance using specific retrieval and generation metrics.
Highlights
- Evaluation-First Design: Built-in tracking for retrieval accuracy, generation quality, and latency.
- Easy Integration: Includes a drop-in npm package (
simba-chat-widget) for instant website deployment. - Highly Customizable: Supports swapping LLMs, vector stores, embedding models, and rerankers.
- Production-Ready: Features streaming responses, async processing, and a management dashboard for documents and conversations.