Skip to main content
Buildactive

Kuni

Python library for processing medical notes with multimodal embedding, clustering, and vector search. Supports text and vision embeddings, zero-shot classification, and clinical assessment processing.

PythonAINLPMedicalComputer Vision
Year2025RoleSole creatorOutcome298 commits across kuni ecosystem

Overview

Kuni is a Python library for processing unstructured medical notes with advanced multimodal embedding, clustering, and vector search capabilities. Optimized for early childhood disability clinical documentation (ages 0-8).

Capabilities

  • Text embeddings — MiniLM for lightweight prototyping
  • Multimodal embeddings — Qwen3-VL-8B for production (text, image, video)
  • Zero-shot classification — Clinical assessment identification across 63+ assessments
  • HDBSCAN clustering — With UMAP dimensionality reduction
  • Vector search — Qdrant integration for semantic retrieval
  • Cross-encoder reranking — For precision ranking of results

Ecosystem

The Kuni Plugin provides an AI agent interface layer with 5 specialist agents for clinical document analysis. Together they form a full-stack clinical intelligence system.

Part of the SSEC Research Software Engineering research portfolio.