Archive
AWS
+2
Jul 9, 2026
•
9 min read
SL#95 - AWS AI Series (28/30) - GenAI Observability End to End: Ship a CloudWatch Dashboard and a Quality-Regression Alarm for Your Agent
Your agent runs green while returning wrong answers. We wire an AWS Strands agent into CloudWatch GenAI Observability, then add the one signal the built-in metrics cannot give you: an online eval that scores live responses and fires a CloudWatch alarm when quality drops. Plan for about two hours.
AWS
+2
Jul 8, 2026
•
8 min read
SL#94 - AWS AI Series (27/30) - When Bedrock Isn't Enough: Deploy an Open-Weights Model on SageMaker AI With JumpStart and Autoscaling
Deploy an open-weights LLM to a real-time SageMaker endpoint in a few lines with JumpStart, put a target-tracking autoscaling policy in front of it, and get a decision table for when to reach for this instead of Bedrock.
AWS
+2
Jul 6, 2026
•
9 min read
SL#92 - AWS AI Series (26/30) - S3 Vectors vs OpenSearch: Build the Tiered Setup and Do the Latency-Cost Math
We'll put a corpus into Amazon S3 Vectors, benchmark its real query latency, write the cost model that tells you when to keep vectors cheap versus when to promote them to OpenSearch, then export the hot subset to an OpenSearch Serverless collection. Plan for about 90 minutes.
AWS
+2
Jul 5, 2026
•
8 min read
SL#91 - AWS AI Series (25/30) - Real-Time RAG: Stream Documents Through Kinesis Into an S3 Vectors Index That's Searchable in Seconds
Build a streaming pipeline where a document dropped onto a Kinesis stream is embedded, indexed in S3 Vectors, and answerable by a RAG query within seconds. No search cluster required.
AWS
+2
Jul 4, 2026
•
9 min read
SL#90 - AWS AI Series (24/30) - Zero-ETL in Practice: Query Live DynamoDB Data in Redshift With No Pipeline Code
Wire a DynamoDB table to a Redshift Serverless workgroup so operational data lands in your warehouse automatically, then query it with SQL. All of it from boto3 and the Redshift Data API. Plan for 90 minutes, most of it waiting on AWS to provision.
AWS
+2
Jul 3, 2026
•
11 min read
SL#88 - AWS AI Series (23/30) - Clean Your RAG Corpus Before It Poisons Your Answers: A Glue 5.1 Dedup and Quality Gate
Build a serverless Glue 5.1 job that strips junk, removes exact and near-duplicate chunks, and refuses to write the output unless it passes a Data Quality ruleset.