The wine track: applied AI from vineyard to cellar — predicting ripeness and harvest date, forecasting yield, steering fermentation, catching faults early, optimising blends, and reading the vineyard with computer vision. Grounded in real oenology and viticulture, and honest about where the models help and where the winemaker's judgement still rules. The full track, in order:
-
A Vineyard Yield and Harvest Dashboard in Tableau
Build a Tableau vineyard dashboard mapping blocks, tracking Brix and acidity ripeness curves, and sequencing harvest with Pulse digests.
-
A Wine Fermentation Monitoring Dashboard in Tableau
Build a Tableau fermentation dashboard tracking per-tank Brix and temperature curves, stuck-ferment flags and YAN, with Explain Data diagnostics.
-
A Cellar and Barrel-Ageing Inventory Dashboard in Tableau
Build a Tableau cellar dashboard tracking barrels by type and age via LOD, top-up and evaporation, SO2 status and an ageing timeline with Pulse alerts.
-
A Wine Tasting-Score and Blending Dashboard in Tableau
Build a Tableau blending dashboard with bench-trial scores by lot, attribute radar charts and what-if candidate blends via parameter actions.
-
A Wine DTC and Varietal Sales Dashboard in Tableau
Build a Tableau wine sales dashboard comparing DTC and distribution, sales by varietal, region and vintage, with club retention and a market map.
-
Predicting Grape Ripeness and the Optimal Harvest Date
How AI models grape ripeness from Brix, acidity, pH, phenolics, and weather to forecast the optimal harvest date and where the winemaker's palate still wins.
-
AI for Vineyard Yield Forecasting
How AI forecasts vineyard yield from bunch counts, berry weight, NDVI remote sensing, and weather (to plan tanks, barrels, and sales) and why vintage weather still breaks the model.
-
Wine Tasting Meets the Data Stack: AI, Power BI, and ERP
Tie tasting and bench-trial scores to vintage, block, lot, and barrel in ERP, analyse in Power BI, and use AI to predict and flag faults: palate still decides.
-
AI for Wine Fermentation Control (and Stuck Ferments)
How AI models wine fermentation kinetics and stuck-ferment risk from YAN, sugar, and temperature, to guide nutrient and temperature decisions for whites and reds, and manage MLF.
-
Detecting Wine Faults Early: Brett, Volatile Acidity, TCA
How AI flags wine faults early (Brettanomyces, volatile acidity, and TCA cork taint) from cellar conditions and rapid analysis, since most faults are easier to prevent than reverse.
-
AI for Wine Blending Optimisation
How AI treats wine blending as constrained optimisation (matching lots, varieties, and barrels to a target style at best use of premium parcels) while the winemaker's palate stays the arbiter.
-
Computer Vision for Grape Sorting and Vineyard Disease Detection
How computer vision sorts grapes at the crush pad and spots vineyard disease from drone and phone imagery, faster and more consistent than manual, with honest limits on lighting and edge cases.
-
Databricks for Wineries: 20 Use Cases
How a winery uses Databricks end to end (ingestion, real-time monitoring, the Delta Lakehouse & Spark, BI and AI) across 20 concrete use cases grouped by capability.
-
Databricks Across the Winery Business, Vertical by Vertical
A department-by-department tour of where Databricks helps a winery (from the floor through quality, supply chain, sales, marketing, finance and compliance) on one governed platform.
-
Winery Energy: AI for Cooling, Presses and Peak Demand
Winery energy spikes at harvest with refrigeration and presses. How data and AI cut power and peak demand (load forecasting, setpoint optimisation and demand management) across regions.
-
Snowflake for Wineries: 20 Use Cases
How a winery uses Snowflake end to end (ingestion, real-time monitoring, Dynamic Tables & Snowpark, BI and AI) across 20 concrete use cases grouped by capability.
-
How a Winery Should Start with AI and Gen AI: The Phases
A phased roadmap for a winery adopting AI and generative AI (from collecting data to dashboards, predictive models, GenAI copilots and agents) with what to do, need and watch at each phase.
-
Snowflake Across the Winery Business, Vertical by Vertical
A department-by-department tour of where Snowflake helps a winery (from the floor through quality, supply chain, sales, marketing, finance and compliance) on one governed platform.
-
Claude AI and Claude Code for Wineries: Where the Anthropic Ecosystem Helps
Where Claude, Claude Code, the API, agents and MCP help a winery (vineyard and viticulture, winemaking, lab, sales, marketing, compliance and the wine club) and where a human must stay in the loop.
-
Microsoft Fabric for Wineries: 20 Use Cases (and 3 Case Studies)
How a winery uses Microsoft Fabric (OneLake, Real-Time Intelligence, Lakehouse, Direct Lake and Copilot) across 20 use cases from vineyard and fermentation to barrel ageing and DTC, plus three case studies.
-
IoT in the Winery: Sensors from Vineyard to Bottle
A process-grounded guide to IoT in winemaking: sensors in the vineyard, at crush and fermentation, in the barrel room and at bottling, the edge-to-cloud stack, and the AI on the streams.
-
Descriptive, Diagnostic, Predictive, Prescriptive: The Four Analytics in a Winery
The four classic types of data analytics translated into winemaking language (yield per hectare, fermentation curves, wine faults, blend trials, barrel ageing) and where Microsoft Fabric's lakehouse, real-time intelligence and Direct Lake actually fit each one.
-
Why Power BI Dashboards Die in Wineries: The Domain Gap
A skilled Power BI developer walks into a winery and a vintage later the dashboards are abandoned. Not a tooling failure but a translation failure: units, lot genealogy, vintage time and KPIs that don't match how winemakers actually think.
-
The Translator: When the Winemaker Is Also the Power BI Developer
What changes when one person has stood on the crush pad, run the ferments and pulled barrel samples and also holds the Power BI and Microsoft Fabric certifications. Why the hybrid profile builds winery dashboards that survive a vintage, and the honest limits of being one.
-
Ask the Chatbot for the Alcohol, Get Three Answers: GenAI Needs a Semantic Layer
Part 1 of The Cellar Ledger. A chat-with-your-data assistant asked for potential alcohol at a winery returns three different numbers, because the conversion lives in three spreadsheets. Why text-to-SQL agents amplify undefined metrics, and how a governed semantic layer, instrument-aware pipelines and a small eval set fix it.
-
Event-Sourcing the Cellar: Why Winery Volumes Should Be Derived, Never Stored
Part 2 of The Cellar Ledger. When tank volumes are an editable field, losses disappear without a reason. An append-only movement ledger in the lakehouse makes every balance a calculation, gives an AI agent real provenance to explain, and keeps the agent from ever overwriting the truth.
-
The Loss Map: A Medallion Pipeline, Anomaly Flags and an LLM That Drafts the Variance Note
Part 3 of The Cellar Ledger. Every winery loses wine between crush and bottle. Build the loss map as a bronze, silver and gold pipeline over the movement ledger, flag unusual stages with robust statistics, and let an LLM draft the monthly variance note without ever doing the arithmetic.
-
Lot Genealogy as a Graph: Recursive Queries and a Recall Agent Built on MCP Tools
Part 4 of The Cellar Ledger. Which bottles contain fruit from block 7, and how much? Model the cellar as a graph of lots and volume-weighted movements, trace it with a recursive query, and put a GenAI agent in front of it through MCP tools that do the maths so the model does not have to.
-
Excise Returns as Data Contracts: Tests Reconcile, the LLM Drafts, a Person Signs
Part 5 of The Cellar Ledger. A winery excise return is a balance equation per tax class. Treat its inputs as a data contract, test the equation in the pipeline every night, and use GenAI for what it is good at here: classifying messy removal notes, drafting the explanations and finding the right paragraph of guidance.
-
Where Winery AI Breaks: Ten Vintages Is Ten Rows
Part 6 of The Cellar Ledger. A brewery brews hundreds of batches a year. A winery gets one vintage. Why models that predict vintage outcomes overfit, where winery data is actually rich, and what time-series foundation models, synthetic data and LLMs can and cannot do about a small-data problem.
-
Whose Palate Did the Model Learn? Bias in Sensory AI, and Evals That Catch Hallucinated Tasting Notes
Part 5 of Responsible AI for Beer, Whisky and Wine. Flavour models and LLMs learn a Western tasting vocabulary and a narrow panel, then write confident notes that invent casks, vineyards and awards. How to find the bias in descriptor data and panels, how to ground note drafting in the batch record, and how to build an eval set that checks every claim in a tasting note against its source.