junruxiong/IncarnaMind

Implements a Sliding Window Chunking mechanism that dynamically adapts chunk size based on data complexity, paired with an Ensemble Retriever to balance fine-grained and semantic retrieval across document collections. Built on LangChain and Chroma DB, it enables multi-hop queries spanning multiple documents while maintaining compatibility with quantized GGUF models (via llama-cpp) for local inference, alongside commercial APIs from OpenAI, Anthropic, and Together.ai.

799 stars. No commits in the last 6 months.