Machine Learning Competency
Submitted for applied AI that acts inside a product rather than alongside it — agent tooling, evaluation harnesses, and inference cost discipline.
- Under review
- July 2026
- Expected Q4 2026
- 2 submitted
What it covers
Agents that act, not sidebars that suggest.
Our submission centres on production agentic systems — where the model invokes the same capabilities the interface does, and every action is auditable.
Agent tooling
Shared capability registries called by both the model and the UI.
Evaluation harnesses
Measurable agent behaviour, so model changes are testable rather than felt.
Inference discipline
Model selection and routing tuned against real cost and latency budgets.
Auditable actions
Every agent-invoked call recorded against the workspace history.
Validation criteria
What AWS assessed.
This competency is under AWS review. The criteria below reflect what has been submitted and what remains outstanding — we would rather show the real state than imply a badge we do not yet hold.
How validation worksProduction workloads
Two AI-native products running on Bedrock with real users.
Customer references
Seshira and Mercado submitted with architecture detail.
Technical review
AWS solutions architect review scheduled for Q4.
Final validation
Decision expected before the end of the year.
Evidence
The engagements behind it.
These are the customer references submitted to AWS for this competency. Ask us about any of them.
24
Executable agent tools
Seshira — shared tool registry
One capability registry called by both the agent and the interface, with every invocation auditable.
Native order channel
Mercado — AI order parsing
Free-form chat messages converted into structured, priced order lines that land reconciled.
Architecture
Patterns behind this practice.
Services
What we build it on.
- Bedrock
- Bedrock AgentCore
- SageMaker
- Lambda
- DynamoDB
- EventBridge
Team on this practice
Applied AI runs out of the engineering bench, with the practice lead who built Seshira's agent layer.
Other competencies.
All designationsMigration & Modernization
Validated for moving production estates onto AWS without a maintenance window, and for decom…
Cloud Operations
Validated for operating what we build — observability, incident practice, cost management, a…
Data & Analytics
Validated for warehouses, streaming pipelines, and managed ETL that stay operable after the …
Security
Submitted for residency-first architecture, least-privilege identity, and audit trails desig…
Work with this practice
Building AI that actually does something?
Discovery comes first — a short, bounded review that ends in an architecture, a scope, and a price. You keep the output either way.