Kumo

[Kumo](http://www.kumo.ai/) is a Snowflake Native App that helps data scientists make highly accurate predictions about user and customer behavior by combining graph learning over structured data and generative AI models trained for unstructured data - all entirely within Snowflake. Named Forbes AI 50 in 2024, Kumo’s predictive AI learns across multiple Snowflake tables, to build accurate predictive machine learning (ML) models in hours - not months. The out of the box machine learning models can be refined for maximum performance and predictions can be delivered in batches or as embeddings for use downstream.  Use a low-code interface to simplify MLOps, combine graph learning and LLMs, and use Kumo to accurately predict the right products, actions, and offers for each person to drive improvements to KPIs like conversion, engagement, and retention.

<p><br/></p>

**Product Highlights:**

- Graph transformers eliminate feature engineering and create predictive AI models from your Snowflake data
- Open-source and commercial LLMs like Cortex unlock unstructured data for even better predictions
- Graph neural networks and LLMs bring deep learning inside Snowflake to better use tabular, relational data for predictive AI
- Kumo’s AI learns directly on multiple Snowflake tables, and writes back predictions as new Snowflake tables
- Automatically retrain models based on the latest data and serve predictions directly from Snowflake tables in production

<p><br/></p>

**Build predictive AI with all Snowflake data**

- Automatically represent your Snowflake data as a richly labeled Kumo graph stored in Snowflake
- Kumo uses AI and LLMs to learn from the graph to produce accurate predictions
- Code and Snowflake workloads like ETL and EDA can be reused in Kumo

<p><br/></p>

**Improve predictive model performance and prediction accuracy**

- A single Kumo graph supports many predictive models and use cases without redesign
- Complex predictive tasks models can be built in fewer than 20 lines of code
- Model architecture and other learning parameters can be tuned by autoML or adjusted manually

<p><br/></p>

**Go from predictions to production**

- Embeddings and predictions are stored as new Snowflake tables
- Models can be automatically retrained and predictions refreshed at your desired interval
- Add predictions and embeddings to existing ML pipelines and dataflows

<p><br/></p>

**Kumo’s Native App in Snowflake Marketplace:**

- Authenticate using Snowflake Single Sign-On
- Data never leaves leaves the customer's Snowflake environment 
- Data and artifacts generated by Kumo are stored in the customer’s Snowflake environment and inherit Snowflake retention policies
- Kumo inherits all Snowflake security and governance capabilities 
- Seamlessly integrate with existing Snowflake workflows and policies
- Use Snowflake credits to pay for Kumo

<p><br/></p>

**Explainable AI**

- Kumo provides detailed model evaluation, model explainability, MLOps and other capabilities needed for building trust and monitoring models in production 
- Get detailed explanations, generated through backpropagation of gradients, all the way down to the raw data 
- Better understand: column effect for everything prediction, exactly how different inputs contributed to a specific prediction, and relevant trends hidden in raw input data

## Resources

Kumo website: [www.kumo.ai](http://www.kumo.ai)

Kumo documentation: [https://docs.kumo.ai/](https://docs.kumo.ai/)

搜集汇总

背景与挑战

背景概述

Kumo是一款Snowflake原生应用，通过整合图学习和生成式AI模型，帮助数据科学家在Snowflake环境中快速构建高精度的预测模型，并简化MLOps流程。它利用图神经网络和LLM处理结构化与非结构化数据，支持自动训练和预测，所有数据和结果均存储在Snowflake平台内，确保安全性与集成性。该应用还提供可解释的AI功能，以增强模型信任度。