Modern supply chain operations face a critical challenge: efficiently managing raw material inventory across multiple manufacturing facilities. Operations managers must constantly balance inventory levels, deciding whether to transfer materials between plants with excess and shortage, or purchase new materials from suppliers. Making these decisions manually is time-consuming, error-prone, and often results in suboptimal cost outcomes.

This quickstart demonstrates how to build an intelligent supply chain assistant using Snowflake Intelligence and Cortex AI capabilities. By combining natural language querying with semantic search over both structured and unstructured data, you'll create a complete solution that helps operations managers make data-driven decisions about inventory management.

Supply chain operations managers face daily challenges managing raw material inventory across manufacturing facilities:

This solution leverages Snowflake Intelligence and Cortex AI capabilities to create an intelligent assistant that:

1. Answers Ad-Hoc Questions: Operations managers can ask natural language questions about inventory levels, orders, shipments, and supplier information - the agent automatically converts questions to SQL and executes them
2. Provides Contextual Information: The assistant can search and retrieve relevant information from supply chain documentation using semantic search
3. Intelligent Routing: Automatically determines whether to query structured data (via Cortex Analyst) or search documents (via Cortex Search) based on the nature of the question
4. Complex Analysis: Handles sophisticated multi-table queries like identifying plants with low inventory alongside plants with excess inventory of the same materials, and comparing costs between suppliers and inter-plant transfers
5. No-Code Agent Creation: Build and deploy the entire solution using Snowflake Intelligence's visual interface without writing application code

Snowflake Cortex provides fully managed Generative AI capabilities that run securely within your Snowflake environment and governance boundary. Key features include:

Cortex Analyst - Enables business users to ask questions about structured data in natural language. It uses a semantic model to understand your data and generates accurate SQL queries automatically.

The semantic model definitions you create are automatically translated into SQL code that powers the natural language queries. For example, defining business concepts like "Low Inventory" and "Excess Inventory" in the semantic model results in precise SQL WHERE clauses:

Cortex Search - Provides easy-to-use semantic search over unstructured data. It handles document chunking, embedding generation, and retrieval, making it simple to implement RAG (Retrieval Augmented Generation) patterns.

Cortex Agents - Orchestrates multiple AI capabilities (like Analyst and Search) to intelligently route user queries to the appropriate service and synthesize responses.

Learn more about Snowflake Cortex.

Snowflake Intelligence is a unified experience for building and deploying AI agents within Snowflake. It provides:

Here's what Snowflake Intelligence looks like in action for supply chain operations. Users can ask natural language questions about their data, and the agent automatically generates SQL queries, executes them, and returns results:

Learn more about Snowflake Intelligence.

In this step, you'll create the complete supply chain database infrastructure with all necessary tables, stages, and sample data.

The database will model a multi-tier supply chain network with suppliers, manufacturing plants, and customers. The demo considers key operational factors including inventory levels, demand, safety stock, lead times with variability, and costs. The intelligent assistant will help answer critical questions like "When do I order more material vs. transfer from another facility?"

1. Navigate to Projects > Workspaces to create a new private workspace in Snowsight
2. Add a new SQL file to your workspace
3. Import the scripts/setup.sql file from the quickstart repository
4. Click Run All to execute the entire script

This script will create:

The database schema models a complete supply chain network with:

Here's the complete Entity Relationship Diagram (ERD) showing all tables and their relationships:

Within the first step, all objects have been created in SUPPLY_CHAIN_ASSISTANT_DB.ENTITIES database/schema. Now you'll upload the necessary files to the internal stages.

1. Navigate to Database Explorer from the left side menu under Catalog
2. Navigate to the SUPPLY_CHAIN_ASSISTANT_DB.ENTITIES database/schema
3. Click on Stages to view available stages
4. Select the SUPPLY_CHAIN_ASSISTANT_PDF_STAGE stage
5. Click the + Files button on the top right
6. Upload the pdfs/Supply Chain Network Overview.pdf file from the quickstart repository

1. In the same Stages view, select the SEMANTIC_MODELS_STAGE stage
2. Click the + Files button
3. Upload both semantic model files:
   • scripts/semantic_models/SUPPLY_CHAIN_ASSISTANT_MODEL.yaml
   • scripts/semantic_models/WEATHER_FORECAST.yaml

These semantic models define the structure, relationships, and verified queries that Cortex Analyst will use to answer natural language questions about your supply chain and weather data.

You have two options to create the Cortex Search service that will enable semantic search over your supply chain documentation.

1. In your workspace in Snowsight, add a new SQL file
2. Import the scripts/configure_search_services.sql file from the quickstart repository
3. Click Run All to execute the script

This script will:

The search service will automatically index the parsed and chunked PDF content, making it available for semantic search queries.

If you prefer to create the search service manually through the UI:

Even with the UI approach, you'll need to run SQL to parse and prepare the documents. In your workspace, add a new SQL file and run:

1. In Snowsight, navigate to AI & ML > Cortex Search in the left navigation
2. Click Create button
3. Configure the search service:
   • Database: SUPPLY_CHAIN_ASSISTANT_DB
   • Schema: ENTITIES
   • Name: SUPPLY_CHAIN_INFO
   • Source Table to be Indexed: PARSED_PDFS
   • Search Column: Select PAGE_CONTENT
   • Attributes: Click Next
   • Select Columns: Select TITLE
   • Target Lag: 1 hour
   • Warehouse: SUPPLY_CHAIN_ASSISTANT_WH
4. Click Create Search Service

The search service will begin indexing your parsed PDF content and will be ready to use once indexing completes.

Now that you have your semantic models and search service created, you can combine them into an intelligent agent using Snowflake Intelligence. The agent will intelligently route user questions to the appropriate tool based on the nature of the query.

1. Click on Agents within the AI & ML section on the left-hand navigation bar in Snowsight
2. Click Create Agent button
3. • Database SNOWFLAKE_INTELLIEGENCE
4. • Schema AGENTS
5. Agent object name Supply_Chain_Agent
6. Display name Supply Chain Agent
7. Once created, navigate to Tools tab

1. Click + Add next to Cortex Analyst
2. Configure the tool:
   • Select the Semantic model file radio button
   • Database SUPPLY_CHAIN_ASSISTANT_DB
   • Schema ENTITIES
   • Stage SEMANTIC_MODELS_STAGE
   • Select SUPPLY_CHAIN_ASSISTANT_MODEL.yaml
   • Name: SUPPLY_CHAIN_ASSISTANT_MODEL
   • Description: "Tool for analyzing supply chain data."
   • Warehouse: Select Custom radio button then choose SUPPLY_CHAIN_ASSISTANT_WH
3. Click Save

1. Click + Add next to Cortex Analyst
2. Configure the tool:
   • Select the Semantic model file radio button
   • Database SUPPLY_CHAIN_ASSISTANT_DB
   • Schema ENTITIES
   • Stage SEMANTIC_MODELS_STAGE
   • Select WEATHER_FORECAST.yaml
   • Name: WEATHER_FORECAST
   • Description: "Tool for analyzing weather data."
   • Warehouse: Select Custom radio button then choose SUPPLY_CHAIN_ASSISTANT_WH
3. Click Save

1. Click + Add next to Cortex Search Services
2. Configure the tool:
   • Database SUPPLY_CHAIN_ASSISTANT_DB
   • Schema ENTITIES
   • Search Service: Select SUPPLY_CHAIN_ASSISTANT_DB.ENTITIES.SUPPLY_CHAIN_INFO
   • Name: SUPPLY_CHAIN_INFO
   • Description: "Tool for searching supply chain unstructured data."
   • ID Column: PAGE_URL
   • Title Column: TITLE
3. Click Save

For each of the following custom tools, click + Add next to Custom Tool, then configure:

1. Click Save to save your agent configuration
2. You can test the agent directly in the right-hand pane
3. Alternatively, navigate to Snowflake Intelligence in the left navigation AI & ML menu
4. Select your Supply_Chain_Agent from the dropdown
5. Start asking questions!

We are building for three core user archetypes involved in supply chain operations.

Role: Manages day-to-day operations and material flow. Responsible for resolving inventory issues, coordinating logistics, and keeping the supply chain moving.

Goals: Maximize asset availability, resolve stockouts before they impact production, and optimize for cost and speed.

Needs: Real-time, actionable answers. A prioritized list of critical issues. Tools to quickly compare options and take action.

Frustrations: Constant "fire-fighting." Waiting hours or days for an analyst to run a report. Making decisions with incomplete data (e.g., not knowing about weather disruptions).

Role: Oversees supply chain budget and long-term strategy. Responsible for profitability, efficiency, and risk management.

Goals: Ensure operational continuity, meet financial targets, and understand high-level risks and opportunities.

Needs: A high-level, aggregate view of supply chain health. Clear, concise summaries of critical problems, their financial impact, and the recommended solution.

Frustrations: Unforeseen budget overruns. Surprises from major disruptions. Reports that are too detailed and don't provide a clear recommendation.

Role: The technical and business-logic expert. Responsible for data quality and enabling the AI agent by teaching it the business's unique logic.

Goals: Ensure the agent's answers are accurate and trusted. Codify business definitions. Reduce his own ad-hoc reporting backlog.

Needs: A simple UI to define business semantics. The ability to test, validate, and "bless" queries. Tools to debug why an agent gave a specific answer.

Frustrations: Vague requests from users. The same simple questions being asked repeatedly. Business users misinterpreting raw data.

As Maria (The Operator), I want to ask, "Where do I have critical low inventory levels?" and instantly see the most critical plants, so I can immediately know where to focus.

When Maria discovers the Miami plant is critically low on rare earth materials, she asks, "For plants with low inventory of a raw material, compare the cost of replenishing from a supplier vs transferring from another plant with excess inventory." The agent compares transfer from Austin (71% savings) vs. purchasing from a supplier, so she can make the most cost-effective decision.

Before finalizing the transfer, Maria asks, "What type of weather events might impact this transfer?" The agent automatically checks the Weather Forecast tool and warns: "A major hurricane is forecasted; transport may be impossible," so she doesn't make a costly logistical mistake.

As Victor (The Executive), I want to receive a clear summary when Maria asks the agent to "Draft an executive summary email with this analysis, our options, and a recommendation." The agent generates an HTML-formatted email with Executive Summary, Financial Comparison, Key Risks, and Final Recommendation, so I can make an informed decision and reply "Approved" in minutes.

As Leo (The Data Steward), I previously defined business logic like "critical low inventory" in the semantic model YAML file and added verified queries for cost comparisons, so the agent understands our business rules and Maria can trust the answers.

Now it's your turn to experience the Supply Chain Assistant! Follow Maria's journey by asking these questions in order, or explore on your own. Notice how the agent automatically determines which tool to use based on your question!

Ask these questions in sequence to replicate the scenario above:

1. "Where do I have critical low inventory levels?"
2. "Where do we have low inventory of rare earth materials?"
3. "For plants with low inventory of a raw material, compare the cost of replenishing from a supplier vs transferring from another plant with excess inventory."
4. "What type of weather events might impact this transfer?"
5. "Draft an executive summary email with this analysis, our options, and a recommendation."

Try these variations to explore different capabilities:

Congratulations! You've built a comprehensive Supply Chain Assistant powered by Snowflake Intelligence that combines multiple AI capabilities into a single, intelligent agent.

Your Supply Chain Assistant now includes:

As your organization's needs grow, you can evolve from a single agent with multiple tools to a multi-agent architecture where a primary orchestrator agent coordinates with specialized domain agents. This architecture enables better scalability, domain expertise, and independent development of agents for different business functions:

Access the complete code, scripts, and files for this quickstart: