Convey structured and unstructured information along with Amazon Fast

Organizations accumulate information in lots of kinds. Structured information lives in databases and spreadsheets, whereas unstructured information sits in paperwork, stories, and information saved throughout file methods. Analyzing each collectively usually requires switching between a number of instruments, operating separate queries, and manually correlating outcomes.

Amazon Fast addresses this with a unified expertise. You may join structured information by enriched datasets and unstructured information by S3-backed data bases. You may then question each by a single conversational AI agent.

On this submit, you discover ways to:

• Join Amazon Easy Storage Service (Amazon S3) as an information supply and create a data base.
• Construct a House that mixes structured and unstructured information.
• Create and configure a customized chat agent that queries throughout each information varieties.
• Ask questions that span structured and unstructured sources in a single dialog.

Resolution structure

The next diagram illustrates how every element connects to ship a unified information expertise by Amazon Fast.

The structure has three layers:

• Information layer: Your information exists in two kinds. Unstructured content material (paperwork, PDFs, stories) is saved in Amazon S3 and listed by a data base. Structured information (metrics, gross sales figures, and operational key efficiency indicators) is loaded as enriched datasets with semantic descriptions that the agent makes use of to translate questions into SQL.
• Workspace layer: A House brings each information varieties collectively right into a single workspace. The House references your data base for unstructured queries and your datasets for structured queries. It may well additionally embrace dashboards, file uploads, and motion connectors. That is the layer the place information from totally different sources converges right into a unified context.
• Agent layer: The customized chat agent hyperlinks to the House and makes use of it as its data supply. When a person asks a query, the agent determines whether or not to question the data base, the dataset, or each, and returns a mixed response with citations. Amazon Fast additionally applies built-in AI guardrails, akin to content material security screening, to talk interactions.

With this layered strategy, you possibly can develop the answer over time. If you add a brand new information supply to the House, the agent linked to that House can use it, and responses respect the permissions of the person who asks.

Stipulations

To comply with together with this walkthrough, you want:

• An Amazon Fast account with an Enterprise subscription for the person who follows this walkthrough (creating datasets requires Enterprise). For particulars, see Amazon Fast pricing.
• An Amazon S3 bucket for what you are promoting paperwork. To create one, see the Amazon S3 Consumer Information.
• Amazon Fast entry to your S3 bucket, granted by an Amazon Fast administrator. Within the admin console, the administrator chooses Permissions, then AWS assets, selects Amazon S3, and chooses Choose S3 buckets. Chosen buckets have read-only permissions by default.
• Permissions to create integrations, data bases, datasets, areas, and chat brokers in Amazon Fast.

Put together your pattern information

This walkthrough makes use of two kinds of information:

Unstructured paperwork: Add these to your Amazon S3 bucket for the data base:

Structured information: Obtain this file to your pc. You add it on to Amazon Fast in Step 3:

Step 1: Create an Amazon S3 integration

Begin by connecting Amazon Fast to your S3 bucket. From the Amazon Fast console, choose Data within the left navigation panel.

The data web page reveals out there connector varieties, together with Amazon Q Enterprise, Internet Crawler, Amazon S3, and third-party choices like Google Drive, OneDrive, and Microsoft SharePoint.

Below Amazon S3, select Add. Within the Create S3 data base dialog, on the Join S3 bucket web page, present the next particulars:

• Title: A descriptive identify on your integration (for instance, “My Information Entry Integration”).
• S3 bucket location: Select Fast Suite occasion account to make use of a bucket in the identical AWS account as Amazon Fast, or Different AWS account for a bucket in a unique account.
• S3 bucket URL: The S3 URI of the bucket containing your information. The bucket have to be in the identical Area as Amazon Fast.

Select Subsequent.

Determine 2: Create integration dialog exhibiting the S3 connection fields

Step 2: Create a data base

After the S3 integration is created, Amazon Fast prompts you to create a data base. The data base indexes your S3 content material for AI-powered search.

Configure the data base:

• Title: “Enterprise Data Base”.
• Description (optionally available): Notes about how the data base might be used.
• Content material: Choose “Add all content material” to index all the pieces within the related S3 bucket, or “Add particular content material” to specify the S3 prefixes for the folders and information to incorporate (prefix filters are case-sensitive).

Subsequent: Further settings opens optionally available settings for document-level entry management lists (ACLs) and a metadata information folder. Determine whether or not to make use of ACLs earlier than you create the data base, as a result of you possibly can’t change that selection later. This walkthrough makes use of pattern information, so the default settings are ample. For particulars, see Arrange and handle Amazon S3 integration.

Select Create. A hit banner confirms the S3 integration, and the data base begins indexing within the background.

Determine 3: Create data base dialog with the Enterprise Data Base identify and content material choices

Step 3: Create and enrich a dataset

A dataset is your structured information file loaded into Amazon Fast. By enriching the dataset with semantic descriptions, customized directions, and proper information varieties, the AI agent can interpret your information extra precisely. It may well then translate pure language questions into SQL queries.

Create the dataset

1. From the Amazon Fast dwelling web page, navigate to Information within the left navigation.
2. Select Create dataset, then choose Add file.
3. Choose the Product_Sales.csv file you downloaded earlier and select Open.
4. On the affirmation display, evaluate the column names and information varieties. Select Subsequent to create the dataset.

Amazon Fast ingests the file into SPICE (the in-memory calculation engine) and the dataset seems in your dataset checklist.

The dataset takes its identify from the uploaded file, so it seems as Product_Sales.csv in your account.

Enrich the dataset

Enriching your dataset with semantic data helps the AI perceive your information in enterprise context. The agent makes use of this context to translate pure language questions into extra correct SQL queries. From the dataset web page, select Edit dataset to enter the info preparation view.

Remodel: Repair information varieties

Within the Remodel tab, replace information varieties for columns that have been auto-detected incorrectly. For this dataset:

• Change Order Date and Date Key from String to Date.
• Change Buyer ID from String to Integer.
• Change Gross sales, Low cost, and Revenue from String to Decimal.
• Change Amount and Unit Worth from String to Integer.

Determine 4: Remodel view exhibiting information sort modifications utilized to the Product_Sales.csv dataset

Output: Add semantic context

Choose the Output tab to entry the dataset-level enrichment fields. Right here you present context that helps the AI perceive what your information represents and the best way to use it.

Description: Add a dataset-level description:

“Product gross sales transaction information for AnyCompany, overlaying orders from FY2024-FY2025 throughout North America, EMEA, Asia Pacific, and Latin America. Contains order particulars, buyer data, product strains, and income metrics.”

Customized directions: Add steering for the way the AI ought to question this information:

USAGE GUIDELINES:

• Income values are in USD.
• When evaluating efficiency, group by Area (North America, EMEA, Asia Pacific, Latin America).
• Order Date needs to be used for time-based development evaluation. The fiscal yr matches the calendar yr.
• Date Secret’s a numeric YYYYMMDD illustration of Order Date.
• Gross sales = Amount x Unit Worth x (1 – Low cost)

COMMON USE CASES:

• Regional gross sales efficiency comparability.
• Product line income evaluation.
• Quarterly development identification.
• Buyer segmentation by business and section.

TAGS: Gross sales, income, orders, regional efficiency, product analytics

Column descriptions: Choose every column within the left panel and add an outline. For instance:

• Order ID: “Distinctive identifier for every gross sales transaction (format: REGION-YYYY-NNNNNN)”
• Area: “Geographic gross sales area: North America, EMEA, Asia Pacific, or Latin America”
• Product: “AnyCompany product identify: Analytics Professional, Join, Automate, AI Assistant, Information Catalog, or Integration Hub”
• Gross sales: “Order income in USD after low cost”
• Section: “Buyer measurement section: Enterprise, Mid-Market, or SMB”

Determine 5: Output view exhibiting dataset particulars with Description and Customized directions configured

Select Save & publish to avoid wasting your modifications. The dataset abstract now reveals the columns with their appropriate information varieties and the descriptions you added.

Determine 6: Enriched Product_Sales.csv dataset abstract exhibiting all 16 columns, information varieties, and descriptions

This enriched dataset offers the structured information layer on your answer. When the agent receives a pure language query about gross sales information, it makes use of these descriptions and directions to generate extra correct SQL queries. Mixed with the data base from Step 2, you now have each structured and unstructured information able to carry collectively in a House.

Step 4: Create a House

Areas in Amazon Fast act as unified information workspaces. A House can maintain file uploads, dashboards, data bases, actions, and datasets, giving your chat agent a single location to question throughout information varieties.

From the left navigation, choose Areas and create a brand new House. Title it “Enterprise Intelligence Hub”.

Add your data base and dataset to the House:

1. Select Add data and choose Data bases from the checklist. Choose “Enterprise Data Base” from the checklist.
2. Select Add data once more and choose Datasets from the checklist. Choose “Product_Sales.csv” from the checklist.

The House now reveals each your data base (unstructured paperwork) and your enriched dataset (structured gross sales information) linked as information sources. That is the central hub the place each information varieties come collectively, giving your chat agent a single place to question throughout all the pieces. Determine 7 reveals the House with each assets within the Prepared state.

Determine 7: Enterprise Intelligence Hub House exhibiting the Product_Sales.csv dataset and Enterprise Data Base

Step 5: Create a customized chat agent

Navigate to Chat brokers within the left navigation and select Create chat agent. Amazon Fast presents a immediate builder the place you possibly can describe your agent in pure language or select from templates like Coaching Information, Undertaking Coordinator, and Onboarding Helper.

For this walkthrough, select Skip to configure the agent manually.

Determine 8: Chat agent creation display with the immediate builder and pattern templates

Configure the agent persona

The configuration panel has a number of sections. Begin with the Agent Persona:

• Agent id: “Enterprise Intelligence Assistant”
• Description: “An AI assistant that breaks down data silos by analyzing enterprise information out of your data base”

For Agent Persona directions, enter:

You're a Enterprise Intelligence Assistant designed to assist organizations break down data silos. Your main features are:

1. Analyze enterprise information saved in S3
2. Retrieve related findings from the stories and memos within the data base
3. Present complete insights that join inside information with exterior tendencies
4. Generate clear, actionable suggestions based mostly on information evaluation

When answering questions, all the time cite your sources and clarify the way you arrived at your conclusions. Deal with delivering sensible enterprise worth and highlighting patterns throughout totally different data sources.

A observe seems confirming that your chat agent comes outfitted with enterprise guardrails.

Determine 9: Agent persona configuration with id and persona directions crammed in

Set communication fashion

Below Communication fashion, choose a response fashion preset. Choices embrace Govt, Technical, and Inventive. For a enterprise intelligence use case, choose Govt.

Hyperlink data sources

Below Data Sources, select Hyperlink areas and choose the “Enterprise Intelligence Hub” House you created earlier. This provides the agent entry to your S3 data base and different information sources in that House. If you hyperlink a House, the agent solutions from massive language mannequin (LLM) data and the linked assets solely.

You can even connect reference paperwork (PDF, TXT, HTML, MD, CSV, DOC, DOCX) on to the agent for added context. Amazon Fast extracts as much as 100,000 characters of textual content from the reference paperwork that you simply add.

Determine 10: Data Sources part exhibiting the linked Enterprise Intelligence Hub House

Customise the agent

Below Customization, configure:

• Welcome message: “Good day! I’m your Enterprise Intelligence Assistant. I will help you analyze your gross sales information and enterprise paperwork and supply insights by evaluating inside data with market tendencies. How can I show you how to at the moment?”
• Steered prompts: “Summarize key insights from our stories” and “Analyze our quarterly gross sales efficiency”

Determine 11: Customization part with the welcome message and steered prompts

Launch the agent

Select Launch chat agent. The precise panel reveals a dwell preview of your agent with the welcome message and steered prompts displayed.

Determine 12: Launched Enterprise Intelligence Assistant with the welcome message and steered prompts within the chat panel

Step 6: Question throughout structured and unstructured information

With the agent operating, begin with a query about what you are promoting outcomes:

How did gross sales change by area in 2025?

For this query, the agent combines outcomes from the Product_Sales.csv dataset with the paperwork within the S3 data base. It returns a response with regional gross sales breakdowns, tendencies, and citations that time again to every supply.

Determine 13 reveals the response. Every quotation marker hyperlinks a determine again to its supply doc.

Determine 13: Chat agent response exhibiting regional gross sales evaluation from the dataset and the data base paperwork

If you ask a query about structured information, the agent makes use of text-to-SQL to translate your pure language right into a SQL question. It runs the question towards the dataset and returns the ends in a readable format. This implies you possibly can ask complicated analytical questions with out writing SQL your self.

Subsequent, ask a few particular doc:

What are the important thing strategic priorities outlined in our product technique memo?

The agent summarizes the strategic pillars, the FY2026 roadmap, and the funding necessities from the Product Technique Memo, as proven in Determine 14.

Determine 14: Chat agent response summarizing strategic priorities from the Product Technique Memo

Lastly, ask a query that wants each information varieties:

How do our 2025 gross sales outcomes align with the market tendencies recognized in our evaluation report?

The agent compares the tendencies within the Market Evaluation Report with outcomes out of your gross sales information and cites every supply, as proven in Determine 15.

Determine 15: Chat agent response evaluating 2025 gross sales outcomes with market tendencies from the Market Evaluation Report

That is the core worth of mixing information varieties in a single House. The agent can pull from structured CSV information and unstructured paperwork in the identical response. You don’t have to know the place the info lives.

Step 7: Generate artifacts and stories utilizing pure language

The Enterprise Intelligence Assistant can even generate formatted paperwork, together with Microsoft PowerPoint displays, Microsoft Excel spreadsheets, and Microsoft Phrase paperwork. Attempt these prompts within the chat agent:

• “Create a PowerPoint presentation summarizing AnyCompany’s This fall 2025 gross sales efficiency by area, together with key tendencies and proposals.”
• “Generate an Excel spreadsheet exhibiting the highest 10 orders by income with Order ID, Area, Product, and Income columns.”
• “Create a Phrase doc with a quarterly enterprise evaluate abstract based mostly on our annual evaluate information.”

Determine 16: Chat agent confirming the request and getting ready a PowerPoint define for AnyCompany’s This fall 2025 gross sales efficiency

Determine 17: Generated eight-slide PowerPoint with a downloadable AnyCompany_FY2025_Sales_Performance.pptx file

Step 8: Share the agent

If you launch a chat agent, it stays non-public till you share it. After you check the agent, share it together with your group:

• From the left navigation, select Chat brokers.
• Within the Actions column for Enterprise Intelligence Assistant, select the menu icon, after which select Share.
• Enter the customers or teams to share with, assign Proprietor or Viewer permissions, after which select Share.

The agent returns permissions-aware responses, and a data base is out there solely to its creator till you replace its permissions. Ensure that the individuals you share the agent with even have entry to the House, the data base, and the dataset. For particulars, see Customized chat brokers.

What you achieved

On this walkthrough, you:

• Linked Amazon S3 as an information supply by an integration.
• Created a data base that indexes unstructured enterprise paperwork.
• Created and enriched a Product Gross sales dataset for structured information queries.
• Constructed a House that serves as a unified information hub on your chat agent.
• Configured a customized chat agent with an outlined persona and linked data sources.
• Queried the agent throughout structured and unstructured information in a single dialog.
• Generated a PowerPoint presentation from a chat immediate.
• Shared the agent together with your group.

Greatest practices

Use the next practices as you progress from this walkthrough to your personal information:

• Arrange your S3 information: Use clear folder buildings and file naming conventions. A transparent construction additionally makes it extra simple to scope a data base to particular S3 prefixes.
• Write particular persona directions: The extra particular your directions, the extra constant the agent responses. Embody steering on the best way to cite sources and deal with ambiguous queries.
• Use Areas to manage scope: Moderately than giving an agent entry to all the pieces, create targeted Areas with solely the info sources related to that agent’s goal.
• Begin with steered prompts: Steered prompts assist customers perceive what the agent can do. Select prompts that exhibit the agent’s vary throughout each structured and unstructured information.
• Check earlier than sharing: Use the preview panel to check your agent with consultant questions earlier than sharing it together with your group.

Safety issues

Below the AWS shared accountability mannequin, you’re chargeable for configuring entry to your information, data bases, and brokers. Contemplate the next as you adapt this answer:

• Grant least-privilege S3 entry: Give Amazon Fast entry solely to the buckets that your data bases want. Buckets that an administrator selects have read-only permissions by default. When you create data bases by the API, you possibly can present a customized AWS Identification and Entry Administration (IAM) function for a person S3 information supply as a substitute of granting account-wide S3 entry.
• Management doc entry: If paperwork require per-user or per-group entry, allow document-level ACLs whenever you create the data base. You may’t change this selection later.
• Encrypted buckets: In case your bucket makes use of AWS Key Administration Service (AWS KMS) keys, comply with the AWS KMS steps within the administrator setup information.
• Add solely trusted content material: The agent makes use of the paperwork in your data base to type responses, so embrace solely content material from sources you belief.
• Share selectively: Chat brokers and data bases are non-public by default. Share them solely with the customers and teams that want them.
• Use guardrails as one layer: Amazon Fast consists of built-in content material security screening, and directors can add as much as 50 blocked phrases or phrases. These controls assist scale back danger however don’t change your accountability to configure entry and permissions. For particulars, see AI guardrails in Amazon Fast.

Clear up

To keep away from ongoing prices, take away the assets you created:

• Delete the chat agent from the Chat brokers web page.
• Delete the House from the Areas web page.
• Delete the Product_Sales.csv dataset from the Datasets web page.
• Delete the data base and S3 integration from the Data web page.
• When you created an S3 bucket particularly for this walkthrough, empty and delete it. Deleting a bucket and its objects is everlasting, so again up any information that you simply wish to preserve.

Amazon Fast person subscriptions and account charges are billed individually from these assets. For particulars, see Amazon Fast pricing.

Conclusion

Amazon Fast helps you carry structured and unstructured information collectively. On this submit, you related S3 as a data base, enriched a dataset, constructed a House, and created a customized chat agent. Enterprise customers can now question each information varieties by a single conversational interface, with no need to know the place the info lives or the best way to write queries.

Concerning the authors