AI/ML: Search and Recommendations (Kendra, Personalize)

Part 3 of 3 in the AWS managed AI/ML series.

These services take longer to provision than Comprehend or Polly. Plan extra wait time after Create.

Kendra availability: AWS announced that Amazon Kendra will not accept new customers after July 30, 2026. Existing customers can continue; new search projects should evaluate alternatives in AWS documentation (for example managed knowledge bases). This page documents Kendra for accounts that already have access.

Prerequisites

aws sts get-caller-identity
export AWS_REGION=us-east-1

Amazon Kendra

Use when: enterprise search over documents in S3, SharePoint, web crawlers, and other connectors—with ranking and FAQ-style answers.

Console

1. Open Amazon Kendra.
2. Indexes → Create index.
3. Index name, description, Create a new role (or pick an existing Kendra index role).
4. Accept defaults on access control unless your lab specifies otherwise → Next.
5. Provisioning → Developer edition for learning → Create.
6. Wait until index status is Active (can take several minutes).
7. Data sources → Add data source → choose Amazon S3 (simplest lab path).
8. Point at a bucket with PDF/HTML/text, sync, then wait for indexing to finish.
9. Use Search in the console (or the test search UI) with a natural-language query.

Reference: Getting started with the Kendra console.

aws kendra create-index \
  --region "$AWS_REGION" \
  --name cli-demo-index \
  --description "CLI demo index" \
  --edition DEVELOPER_EDITION \
  --role-arn arn:aws:iam::ACCOUNT_ID:role/YOUR_KENDRA_ROLE

aws kendra describe-index --region "$AWS_REGION" --id INDEX_ID \
  --query 'Status'

aws kendra create-data-source \
  --region "$AWS_REGION" \
  --index-id INDEX_ID \
  --name cli-s3-source \
  --type S3 \
  --role-arn arn:aws:iam::ACCOUNT_ID:role/YOUR_KENDRA_DS_ROLE \
  --configuration "{\"S3Configuration\":{\"BucketName\":\"YOUR_BUCKET\"}}"

aws kendra start-data-source-sync-job \
  --region "$AWS_REGION" \
  --index-id INDEX_ID \
  --id DATA_SOURCE_ID

aws kendra query \
  --region "$AWS_REGION" \
  --index-id INDEX_ID \
  --query-text "What is our refund policy?"

Amazon Personalize

Use when: item recommendations for users (movies, products, articles) from interaction history CSV in S3.

Flow: dataset group → Interactions dataset → import job → solution (recipe) → campaign → get recommendations.

Console

1. Open Amazon Personalize.
2. Create dataset group → name it → Domain: Custom → Create.
3. Create dataset → Item interactions → attach or create a schema matching your CSV columns (USER_ID, ITEM_ID, TIMESTAMP, etc.).
4. Create dataset import job → select your interactions CSV in S3 → run import until status succeeds.
5. Create solution → pick a recipe (for example User-Personalization for custom domain labs).
6. Wait for solution version status Active (training takes time).
7. Create campaign → select the solution → enable latest solution version → Create campaign.
8. On the campaign page, use Test campaign with a user ID from your CSV.

Reference: Getting started (console).

aws personalize create-dataset-group \
  --region "$AWS_REGION" \
  --name MovieRatingDatasetGroup

# Note dataset group ARN from output, then create schema + dataset + import job
# (see getting-started-cli for schema JSON and create-schema)

aws personalize create-solution \
  --region "$AWS_REGION" \
  --name MovieSolution \
  --dataset-group-arn DATASET_GROUP_ARN \
  --recipe-arn RECIPE_ARN

# After solution version is ACTIVE:
aws personalize create-campaign \
  --region "$AWS_REGION" \
  --name MovieRecommendationCampaign \
  --solution-version-arn SOLUTION_VERSION_ARN \
  --min-provisioned-tps 1

aws personalize-runtime get-recommendations \
  --region "$AWS_REGION" \
  --campaign-arn CAMPAIGN_ARN \
  --user-id USER_ID_FROM_CSV

Troubleshooting

Series complete

Back to Part 1 — Comprehend, Translate, Lex or CLI snippets index.