GenAI architecture
Design and review of LLM and RAG applications on AWS — Bedrock, model selection, retrieval pipelines, serverless backends, and multi-account infrastructure.
AWS Generative-AI Architecture
Build the GenAI system. Then make it affordable to run.
CloudCraft builds LLM and RAG systems on AWS — and makes sure they don’t bankrupt the teams running them. The specialty is GenAI cost and token-efficiency optimization: cutting spend without giving up quality.
Most GenAI systems are architected for one question — does it work? — and then quietly bleed money in production. The skill to build the system and the skill to make it cost-efficient usually live in different people. CloudCraft has both.
CloudCraft is an AWS generative-AI architecture practice. It designs and builds LLM and RAG applications on AWS, and it treats what they cost to run as a first-class part of the architecture — not an afterthought discovered on the first production bill.
What CloudCraft does
Cost & token-efficiency optimization
The specialty. Right-sizing model choice to the task, context and token economics, prompt caching, batch vs. real-time, retrieval cost, embedding strategy, and model routing — cutting spend without giving up quality.
GenAI architecture & cost reviews
A scoped assessment of an existing system: what’s wrong, what it costs, and a prioritized set of fixes — delivered as an artifact your team can act on.
The specialty: cost & token efficiency
A working GenAI system and an affordable one are not the same system. Production spend is set by a dozen architectural choices that are easy to get wrong and hard to unwind later. CloudCraft works the levers that actually move the bill:
- Model right-sizing — matching model choice to the task instead of defaulting to the largest option for everything.
- Context & token economics — controlling what goes into the context window, because tokens are the meter.
- Prompt caching — reusing stable context so it isn’t paid for on every call.
- Batch vs. real-time — routing work that can wait to cheaper asynchronous paths.
- Retrieval cost & embedding strategy — keeping the RAG layer accurate without over-retrieving or over-embedding.
- Model routing — sending easy requests to small models and reserving the expensive ones for where they earn it.
The goal is never the cheapest possible system — it is the same quality for materially less money.
Built, not just advised
CloudCraft’s credibility comes from having designed and operated production AWS systems end to end — Bedrock-backed GenAI, RAG pipelines, serverless backends (Lambda, API Gateway, DynamoDB, Aurora), and multi-account infrastructure as code with CDK — including the production engineering that separates a demo from a system.
CloudCraft’s principal holds the full AWS certification stack — a ‘Golden Jacket’ owner. Golden Jacket owners hold every active AWS certification at once — the complete current stack. It’s a level few ever reach: an estimated few hundred people worldwide have done it. That’s what makes the Golden Jacket such a rare professional distinction — it validates both breadth of expertise across the AWS domain and a sustained commitment to continuous learning. See what CloudCraft has built →
How engagements run
Engagements are deliverable-shaped and run remote and async: scoped, executed, and delivered, with synchronous time kept to a kickoff and a readout rather than a calendar full of standing meetings. That keeps the focus on the work and the artifact — and it tends to move faster. When a live working session is the right tool, CloudCraft uses it.
Running GenAI on AWS — or about to?
A short conversation is the best way to see whether CloudCraft is the right fit for what you’re building or what it’s costing you.