AWS & Amazon Bedrock
We build business platforms on AWS that combine foundation models available through Amazon Bedrock with an organisation's knowledge, tools and workflows.
The foundation
Amazon Bedrock is part of the foundation.
The work
The product experience and systems engineering are Smriti's work.
Platform layers
Choosing a model is only the start. Between the business interface and the selected models sit the workflow logic, the approved data and tools, and the evaluation that shows whether the platform is doing its job.
Amazon Bedrock gives access to foundation models from multiple AI companies through one managed service. The other layers are engineered by us or integrated from the organisation's own systems, with permission.
Business interface
Applications and workflows people use
Agent or workflow logic
Orchestration, rules and tool calls
Selected models
Foundation models available through Amazon Bedrock
Approved data and tools
Knowledge sources and business systems, reached with permission
Evaluation and operations
Measurement, monitoring and improvement
Supporting services
What it provides
Managed retrieval over an organisation's data.
We consider it when
Documents must be searched and cited, and a managed retrieval service suits the data and how it is permissioned.
What we still engineer
Which sources are included, who may see what, how answers cite their sources — and checks on retrieval quality.
What it provides
Configurable safeguards: content filters, denied topics, sensitive-information filters and grounding checks.
We consider it when
An application needs safeguards that can be configured once and applied consistently.
What we still engineer
The policy behind the settings, tests that the safeguards behave as intended, and what the application does when one is triggered.
What it provides
A platform for deploying and operating agents built with any framework and model.
We consider it when
Agents need a managed runtime and operational tooling on AWS.
What we still engineer
The agents themselves: their tools, rules, hand-offs to people and evaluation.
Engineering, not configuration
Model selection
Amazon Bedrock offers foundation models from multiple AI companies. That choice is only useful when it is made against the work the platform must do, rather than a general benchmark.
| Criterion | Model Model A | Model Model B | Model Model C |
|---|---|---|---|
| Task qualityOn the agreed test set | Measured during the project | Measured during the project | Measured during the project |
| Response timeUnder the expected load | Measured during the project | Measured during the project | Measured during the project |
| Cost per completed taskIncluding retries and tool calls | Measured during the project | Measured during the project | Measured during the project |
| Behaviour with safeguardsWith the project's guardrail settings | Measured during the project | Measured during the project | Measured during the project |
| Regional availabilityIn the project's AWS Region | Measured during the project | Measured during the project | Measured during the project |
| Customization, if neededOptions vary by model and Region | Measured during the project | Measured during the project | Measured during the project |
| Decision | Recorded with the trade-offs, and revisited when models change | ||
Capability overview
Business platforms on AWS that combine foundation models available through Amazon Bedrock with an organisation's knowledge, tools and workflows.
A capability overview describes what we engineer in this kind of project. It is not a case study: it names no customers and claims no results.
AWS & Amazon Bedrock