TechPrufer
Get started
Hex
Hex logoHex · Feature trail

Generative Data Apps

14 posts · Jun 2024 – Sep 2026

What this trail covers

Hex has evolved its generative data apps by introducing Generative Chart Controls, allowing users to make precise, point-and-click edits to visualizations within generated apps without needing to prompt an agent. This feature adds a 'Style' button to visualizations, enabling users to modify chart titles, series colors, axis labels, and legend placement directly. This breaks the tradeoff between the flexibility of code generation and the precision of traditional chart controls, enhancing the usability and accuracy of forecasting models by allowing them to continuously learn from recent trends and historical conversion curves. The app now ingests campaign data directly from CRMs, utilizes up-to-date conversion rates, models time-to-convert patterns specific to each pipeline generating activity, and provides a data-warehouse centric approach to forecasting, reducing organizational anxiety and enabling more confident, data-driven decisions.

All Hex trails
  1. You don't need to migrate those 300 dashboardsThis post advocates for migrating the logic and context behind dashboards rather than the UI itself, especially in the context of AI-powered analytics platforms. It suggests using AI agents to translate existing BI dashboard logic into a more accessible 'organizational knowledge' format. The post highlights the value of trusted data and metric definitions over static dashboard UIs, emphasizing that the true goal of self-service is enabling users to ask questions and get trusted answers. It also introduces a 'BI migration skill' that can translate BI dashboards into Hex data apps, reducing the migration effort for valuable dashboards.
  2. AI in business analytics: tools, strategies & case studiesThis post details how AI is reshaping analytics, focusing on three key workflows: conversational analytics, coding agents for exploratory analysis, and AI-generated data apps. It emphasizes the importance of governed context (endorsed tables, business rules, semantic models) for trust and accuracy in AI-generated answers. The post introduces the concept of 'vibe-coded dashboards' and data apps built from natural language descriptions, highlighting the need for these apps to inherit context from the same governed layer as other AI workflows. It also outlines evaluation criteria for AI analytics platforms, focusing on the AI analytics experience, trust and accuracy (inspecting generated SQL/code, using governed definitions, observability), and ecosystem integration across various tools like Slack and ChatGPT.
  3. What are data apps? Types, uses, and how they workThis post introduces the concept of 'data apps' as an evolution beyond traditional dashboards, emphasizing interactive exploration within governed data models. It details how data apps work by connecting directly to cloud data platforms, leveraging warehouse compute, and providing interactive user experiences. The post categorizes data apps into self-service analytics, collaborative exploration tools, and predictive applications, and provides examples of their application across various business functions like finance, sales, and healthcare. It positions Hex as a platform for building these data apps.
  4. How we built a forecasting data app that gets more accurate every quarterThis post details the evolution of Hex's generative data apps, specifically focusing on how a forecasting data app was built and improved over time. It highlights the transition from spreadsheets to a data app, enabling direct CRM ingestion, dynamic conversion rate modeling, and time-to-convert pattern analysis. The app's accuracy has improved quarterly through advanced modeling incorporating segment-specific conversion rates and 90-day conversion windows, and direct CRM integration. Future enhancements explore AI capabilities for generating forecast scenarios and dynamic adjustments. The core technical contribution is the architectural shift to a data-warehouse centric model for forecasting, enabling continuous learning and more accurate predictions.
  5. Forecasting demand to find the perfect inventory balance.This post details how Huckberry used Hex to build a data app for inventory forecasting across 100,000 SKUs, saving over $1M. The app allows business teams to model sales trends, predict unit sell-through, normalize demand curves, and forecast at the SKU level. It provides real-time inventory health tracking up to 26 weeks ahead and generates recommendations for color and size SKUs. The team also built a second app for detailed SKU-level recommendations. This demonstrates the application of generative data apps for complex business problems, enabling scenario modeling and interactive adjustments by business users.
  7. hex cortex overview, all your questions answeredThis post details the integration of Snowflake Cortex LLM Functions within Hex, enabling users to leverage advanced AI capabilities directly within their data workflows. It highlights how Hex simplifies the use of Snowpark and Cortex, offering a "Snowpark Easy Button" for secure session creation and seamless integration of Snowpark dataframes into SQL and Python cells. The post also addresses the use of Cortex LLM Functions for tasks like classifying product recalls and extracting data, discusses considerations for model selection, cost, and context windows, and clarifies the distinction between using managed Snowflake Cortex functions and external APIs like OpenAI. It also touches upon the pushdown compute capabilities of Hex for processing large datasets and the potential for running open-source models in Snowpark Container Services.
  8. Making AI charts go brrrrThis post details the technical optimizations made to accelerate the generation of AI-powered charts in Hex. The primary focus was reducing latency by optimizing the representation of Hex's internal chart spec for LLMs. This involved creating a smaller 'chart stub' that captures 90% of commonly used features with fewer tokens, and developing a custom model fine-tuned to reliably generate this stub. The post also discusses the trade-offs between using native Hex chart cells versus generated Python code, the importance of deep integration, and the use of an internal evaluation framework (Spellgrounds) to test these optimizations.