The use case is concrete.
Users, data sources and the intended outcome are known. The question is how to deploy AI safely, not whether AI is interesting.
All services
AI in production
From AI demo to production system. With access control and a cost ceiling.
An AI demo always works. Production asks the harder questions: who may call which tool on whose behalf? What happens when a call fails? What does one user cost per month? I build the infrastructure behind agentic systems: MCP servers, controlled tool access and LLM integration into the backends you already run.
AWS Certified Generative AI Developer – ProfessionalMCP & tool callingCognito & OAuth 2
What you get
- A feasibility and cost estimate before the build, not after it
- An MCP server with authentication, authorization and an audit trail
- Your existing systems exposed as tools you stay in control of
- Evaluations, guardrails and per-use-case cost monitoring
Scope & working together
The AI production review delivers a concrete implementation plan within ten working days. The production build typically runs two to four months afterwards, with an explicit option to stop after the review. Specific cases run shorter and at an entry price, linked below.
Remote from Germany. Straight with me, no agency in between.
The starting point
The prototype convinces. Operations asks the questions.
An agent that works in a notebook is not a product. The moment real users, real data and real permissions enter the picture, the problem shifts. It stops being about prompts and starts being about identity, access rights, traceability, failure behaviour and cost per request.
The critical part of an agentic system is rarely the model. It is the tools: what an agent may do on behalf of a user, how that is actually enforced, and how you find out afterwards what really happened. That is where I work.
Typical symptoms- The agent runs as a service account that is allowed to do everything.
- Nobody can trace which action an agent triggered.
- Cost per request is unknown and has no upper bound.
- Quality is judged by feel instead of measured.
- Switching models means touching half the application.
- It is unclear which data ends up in a prompt at all.
Who this is for
For teams that must turn an AI use case into a dependable system.
For CTOs, technical leaders and product owners in established companies. A concrete use case or prototype exists; the open questions concern data access, security, cost and operations.
AI must connect to existing systems.
The model needs controlled access to backends, documents or tools and must fit existing identities and workflows.
Operations must be accountable.
Cost, permissions, quality and failure modes need answers before rollout, not after the first incident.
What I do
Controlled.
Traceable.
Affordable.
01
Sharpening the use case
First the uncomfortable question: does this actually need a model? Some of it is a search function, some of it is a rule. If an LLM is the right answer, we define up front what "good enough" means and how it gets measured.
02
The MCP server as a controlled interface
Your existing systems are exposed as tools over MCP, with clearly bounded capabilities. Every tool has a signature, a permission check and defined failure behaviour. No model gets direct database access.
03
Identity and authorization
The agent acts on behalf of one user, not on behalf of everyone. OAuth 2 through Cognito, a token broker for downstream systems, and permissions checked where they belong. Every tool call maps back to a person.
04
Operations, cost and guardrails
Token consumption per user and per use case, hard limits and budgets, caching for repeat requests, smaller models where they suffice. Plus timeouts, retries and a clean fallback when a provider goes down.
05
Making quality measurable
Test sets built from real cases, automated evaluations in the pipeline, regressions visible before release. Changing a model or a prompt becomes a decision backed by numbers instead of a gut call.
Specific cases
AI phone assistantAnswer calls around the clock, handle repeat questions and pass requests to your team as structured data. A dedicated page covering scope, operations and the fixed price of 3,900 euros.Introducing Claude CodeAgents now write the larger part of the code, and output rises immediately. What shows up later is the drift: no single commit is wrong, but after three weeks a module boundary has a silent hole. A page of its own on the structure that prevents it.MCP server developmentAn MCP server is written in an afternoon. What is not written in an afternoon: whose permissions the agent acts under, what the log gives you afterwards, and what a user costs per month. A page of its own on that.Reducing LLM costsAn AWS bill can be broken down, an inference bill knows one total. A page of its own on making cost per request and per user visible, and on which lever comes first.
Want your AI idea in production, secure and affordable?
Let’s spend 30 minutes on the use case.
How it runs
First check whether it is worth it.
Stopping is part of the process. After feasibility you know whether the production build pays off, and you are allowed to say no.
STEP 01
Check the use case
Use case, data situation, expectations. If a model is not the right answer, I say so in the call rather than three months in.
STEP 02
AI production review
A review of architecture, data access, security, costs and integration. The result is a prioritized implementation plan within ten working days.
STEP 03
Production build
MCP server, tool integration, auth and audit trail. Deployed on AWS through Terraform, like any other service.
STEP 04
Operations and handover
Evaluations in the pipeline, cost and quality dashboards, handover to your team including pairing.
Entry offer
AI production review. From 2,400 €
From prototype to secure operations.
I review the architecture, data access, security, cost and integration of your AI solution and deliver a concrete implementation plan.
What you get
- Architecture review from model call through to production operations
- Data flows and access rights: which data goes where and who may trigger what
- Security review covering identity, secrets, tool access and logging
- A cost picture per request and use case, including the main cost drivers
- Review of the integration into existing systems, interfaces and operating processes
- A concrete implementation plan with priorities, risks and the next sensible steps
What you do not get
- No implementation of the recommended measures within the review
- No penetration test and no legal or data-protection assessment
- No blanket approval for production without access to the relevant documentation and systems
- The starting point is an existing prototype or a concretely designed AI solution
- You provide architecture documentation, the relevant access and a technical contact
- The price starts at 2,400 euros net. Scope and binding price are agreed before work starts
- The implementation plan is yours and may be carried out by your team, by me or by another provider
- There is no obligation to commission further work
The outcome
What actually reaches production.
A system that survives an audit
Every tool call has a user, a checked permission and a log entry.
Costs you can plan around
Consumption visible per use case, limits in place, expensive calls caught before they become habit.
Replaceability
Models and providers can be swapped without rebuilding the application.
Technologies I use
Proven tools. No experiments.
AI services
- Amazon Bedrock
- SageMaker
- Claude
- Titan Embeddings
AI building blocks
- MCP
- FastMCP
- RAG
- Tool calling
- Prompt engineering
- Guardrails
Backend
- Golang
- Python
- PHP 8
- TypeScript
- gRPC
- REST
Data
- DynamoDB
- PostgreSQL
- OpenSearch
- S3
- Valkey / Redis
Security
- IAM
- Cognito
- OAuth 2
- KMS
- Secrets Manager
Platform
- ECS Fargate
- Lambda
- EventBridge
- Step Functions
- API Gateway
Infrastructure
- Terraform
- AWS CDK
- Docker
- GitHub Actions
Operations
- CloudWatch
- OpenTelemetry
- Datadog
- Grafana
Who you are talking to
Directly with me as a freelancer. No agency in between.
I am Tim Rutte. More than 20 years in software development. Today I take AI systems into production, where they become measurable. You talk to the person who touches your code, from the first call to the handover.
- 20+years in software development
- 50+successful projects
- 2003working remotely since then
Working together
Working with your team or delivering independently.
You do not need to bring a particular setup. I adapt the engagement to how your company works and how much responsibility you want to hand over.
01Together
I work as part of your team.
If knowledge and responsibilities already sit with you, I join wherever additional experience is needed. Inside your workflows, in direct contact, without creating a parallel track.
- Integration into your sprints, reviews and technical decisions
- Pairing and knowledge transfer throughout delivery
- Code, documentation and operations remain fully with your team
02Independent
I take ownership of delivery.
If you lack time or capacity internally, I take a clearly scoped piece of work from technical clarification through to production. You set the objective and constraints; I take care of the path there.
- One point of contact from clarification through to delivery
- Regular, clear progress updates without day-to-day supervision
- A proper handover with documentation and a walkthrough
Whichever model you choose, you always know what is being built, which decisions are pending and what happens next. Billing follows the project and the model you prefer: clearly bounded work at a fixed price, longer or more open-ended work by time spent.
Continuity
Your system remains yours.
Independent delivery does not create dependency on me. Everything required to develop and operate the result stays in your environment from day one and remains ready for handover throughout the engagement.
Access stays with you.
Code, cloud accounts, pipelines and secrets live in your systems. Operations never depend on a personal account or credentials that only I control.
Knowledge does not stay in my head.
Architecture decisions, operating procedures and known risks are documented where your team will find them and kept current as the work progresses.
Someone else can take over.
Reproducible environments, automated deployments and regular handovers allow your team or another provider to continue without starting again.
The goal is not for you to depend on me permanently. The goal is for you to remain free to decide who develops the system next.
Common questions
Before you ask.
What is an MCP server and why would we need one?
The Model Context Protocol is an open standard for how a model talks to external systems. An MCP server exposes your systems as clearly defined tools, each with a schema, a permission check and defined failure behaviour. The benefit: you control what an agent may do in one place, instead of copying that logic into every application.
Does our data have to go to a model provider?
Not necessarily. Through Amazon Bedrock, calls stay inside your AWS environment and region. Which data reaches a prompt at all is decided at the tool level and logged. That is part of the design, not a setting flipped at the end.
How do we keep costs under control?
Through measurement and limits: token consumption per user and use case, hard caps, caching for repeat requests, and smaller models where they are good enough. The cost estimate comes out of the feasibility phase, before anything is built rather than when the first invoice lands.
We already have a prototype. Can you build on it?
Yes, and that is a common starting point. The prototype usually answers whether it works in principle. What is missing is the path into operations: identity, permissions, failure behaviour, evaluations and cost control. That layer is what I build.
What if the use case turns out not to work?
Then we say so after the feasibility phase and you have spent one or two weeks instead of two quarters. A well-argued no is a legitimate outcome, and considerably cheaper than a project nobody wants to cancel.
Terms that come up here
RAGModel Context ProtocolRate limitingPrompt injectionEmbeddingsGuardrailsLLM evaluationContext windowHallucinationFine-tuningAI agentInference costHuman in the loopVector databaseChunkingRerankingPrompt engineeringSystem promptFunction callingOpen-weight modelsEU AI Act
To the glossaryOther services
What else I help with.
Backend development
Backends for SaaS and platforms that hold under real load: Go and PHP 8, event-driven, with tenant isolation and recovery designed in.
Learn moreLegacy modernization
Modernizing grown systems step by step: strangler fig instead of a rewrite, operations untouched, every step reversible.
Learn moreAWS migration
From your own data centre or another cloud onto AWS, with a cost model before the move and a rollback path for every step.
Learn moreReducing AWS costs
I find where your AWS budget leaks away, and cut it measurably without giving up performance or availability.
Learn moreImproving AWS resilience
So that a single fault does not take the whole application with it: dependencies and single points of failure identified, recovery rehearsed rather than documented, changes delivered as Terraform.
Learn moreProcess automation
Replacing manual workflows with real systems: wired into the software you already run, with permissions and an audit log instead of a chain of tools.
Learn more