DR. Roman Ceresnak

FEATURED SERIES

Production AWS AI Architectures

I Replaced My Custom Agent Orchestration with AWS Strands SDK — Here's What Happened

422 lines of custom ReAct loop deleted. Model-driven loop with 9 DevOps tools, HITL via Step Functions, DynamoDB session memory, and full CloudWatch observability — all on 6 Terraform modules.

✓ @tool auto schema generation

✓ waitForTaskToken HITL pattern

✓ 6 Terraform modules

PageIndex vs Classic RAG: Why Vector Search Fails on Structured Documents

Classic RAG chunks destroy legal document hierarchy. PageIndex builds a tree and lets Bedrock navigate it — achieving 100% cross-reference accuracy with zero embedding API calls at query time.

✓ Zero embedding calls at query time

✓ 100% cross-reference accuracy

✓ Section-level citations

Build Self-Learning Agents With No Fine-Tuning on AWS

Production-grade self-learning agent using Bedrock, Lambda, DynamoDB, and Step Functions. Runtime feedback loop writes new knowledge back to Bedrock Knowledge Base after every interaction.

✓ Self-learning runtime

✓ Zero GPU training

✓ Bedrock Knowledge Base

RAG vs CAG vs KAG: How to Cut AWS Bedrock Token Costs by 50%

Benchmark of three knowledge retrieval patterns — RAG with FAISS, CAG with Bedrock prompt caching, and KAG with Amazon Neptune — showing up to 50% token cost reduction with full Terraform infrastructure.

✓ 50% token cost reduction

✓ 3-pattern benchmark

✓ Full Terraform infrastructure

Explore All AWS / AI Projects →

Hey, I'm Roman

Production AWS AI Architectures

I Replaced My Custom Agent Orchestration with AWS Strands SDK — Here's What Happened

PageIndex vs Classic RAG: Why Vector Search Fails on Structured Documents

Build Self-Learning Agents With No Fine-Tuning on AWS

RAG vs CAG vs KAG: How to Cut AWS Bedrock Token Costs by 50%