Dunieski Otano

04Generative and agentic AI

Agents that make it to production.

RAG assistants, agentic workflows connected to real business systems and the platform underneath them, built on Amazon Bedrock and AgentCore with security and evaluation from day one.

  • Amazon Bedrock
  • AgentCore
  • SageMaker
  • MCP
  • A2A
  • LangGraph
  • Strands Agents

Capabilities

What I build with AI on AWS.

  1. 01

    AgentCore agent platforms

    Runtime, Memory, Gateway, Identity, Policy, Observability and Evaluations, deployed with the AgentCore CLI and CDK from source-controlled configuration.

  2. 02

    RAG assistants on Bedrock

    Knowledge Bases over Amazon S3, chunking and embedding strategies, vector search, then measured on quality, accuracy, latency and cost.

  3. 03

    Integration and orchestration

    AWS Lambda layers that connect models and agents to applications, REST APIs and business systems, with human approval for sensitive actions.

  4. 04

    AI contact center

    Amazon Connect self-service with Amazon Lex and Bedrock, plus call summaries and knowledge answers grounded in customer content.

  5. 05

    Data and model pipelines

    Ingestion, embedding and retrieval pipelines with S3, Glue and Athena, and SageMaker fine-tuning where foundation models fall short.

  6. 06

    Framework flexibility

    LangGraph, Strands Agents, OpenAI Agents SDK and Google ADK on one governed runtime, with MCP and A2A where they fit.

The stack

Everything an agent needs, in orbit around AgentCore.

Amazon Bedrock AgentCore
Amazon Bedrock
Strands Agents
LangGraph
MCP
A2A
AWS Lambda
Amazon Cognito
Amazon CloudWatch
Amazon SageMaker
Microsoft Entra ID
Amazon API Gateway

Reference architecture

How an agent reaches production.

UserApp, chat or voiceAmazon API GatewayThrottling at the edgeAgentCore IdentityTokens and OAuthAgentCore RuntimeIsolated per sessionAgentCore MemoryShort and long termAgentCore GatewayTools over MCPAmazon BedrockFoundation modelsBedrock GuardrailsModel input and outputAgentCore PolicyCedar rules on tool callsAWS LambdaFunctions as toolsMCP serversExisting tool serversREST APIsOpenAPI and SmithyAPI Gateway APIsExisting REST stagesAmazon CognitoUser poolsMicrosoft Entra IDEnterprise sign-inStrands · LangGraphAny agent frameworkAmazon CloudWatchTraces, metrics, logs

Featured work

A hands-on AgentCore enablement program.

A hands-on program that took engineering teams from first agent to production patterns, built around a live concierge agent on AgentCore Runtime and Memory.

  • A live deploy bridge streams real AgentCore CLI events onto an interactive architecture diagram, so the audience watches Runtime and Memory provision and answer in real time.
  • A serverless chat Lambda, runtime status API and Terraform-managed S3 and CloudFront hosting, with agent resources replaceable without infrastructure changes.
  • Memory strategies for semantic facts, user preferences and session summaries, with namespace separation and retention controls.
  • Reusable developer templates, so every new agent starts from a secure, observable, deployable baseline.

Delivery

Discovery to handover, end to end.

Every engagement is measured on quality, accuracy, latency and cost, then handed over with deployment guides and runbooks.

  1. 01

    Discover

    Business problem, data readiness and use-case fit.

  2. 02

    Design

    Solution design, implementation plan and estimate.

  3. 03

    Prove

    Proof of concept with measured quality and cost.

  4. 04

    Ship

    CI/CD, infrastructure as code, guardrails and monitoring.

  5. 05

    Hand over

    Deployment guides and runbooks the team can own.

Contact

Ready to take an agent to production?

contact@otanoacademy.comAll the ways to reach me