Skip to content
RAM CADDSYS
Book Demo

AI for engineering teams

Claude agents for engineering, fabrication and construction firms

RAM CADDSYS has implemented engineering software for 28 years. We now implement Claude the same way: inside the tools and workflows your team already uses, with training for the people who run them.

Discuss a pilot

In production

We run our own company on one intelligence layer

Since 2026 we have connected our own commercial and back-office work to one layer built on Claude. The business can be asked a question, and it can act on the answer. What we implement for you is what we already operate.

  • Sales
  • Conversations
  • Quoting
  • Billing
  • Renewals
  • Finance
  • Hiring
  • Website and SEO
  • Code quality

One intelligence layer

Connected
Each function writes to the layer and reads from it.
Queryable
Ask about a lead, a quote or a renewal and get the answer, with its source.
Closed loop
Scheduled checks watch the work and raise what needs a person.

Services

What we implement

Four kinds of work, usually in this order. Set up Claude where your data lives. Connect the tools your team already runs. Automate the workflow between them. Train the people who own it.

Today, by hand

  1. Result
  2. Export
  3. Email
  4. Re-key
  5. Next tool

Connected

  1. Result
  2. Connector
  3. Claude
  4. Approval
  5. Next tool

Set up

Claude Enterprise setup

Single sign-on and provisioning, groups and roles, the surfaces and connectors each group gets, spend caps and retention.

Deployment on AWS or Google Cloud

Inside your own Amazon Bedrock account or Google Cloud project.

Connect

Your tools, connected

Any tool with an API, a file format, a database or a command line, including the ones you built yourselves. Connected through MCP, so results stop moving by hand.

Design to shop floor

Controller retrofit, CNC programming, and custom control software for machines the manufacturer no longer supports.

Automate

Agents for the commercial desk

Enquiry intake, routing, quote drafting, renewal and purchase-order follow-up.

Knowledge retrieval

Design codes, standards and project documents, answered with source citations.

Enable

Training for every role

The people who use it, the champions who own it, and the engineering and IT teams who extend it with Claude Code.

Enterprise rollout

A written plan: every decision with an owner and a date, then go-live group by group.

Engagement

How an engagement works

One workflow at a time, measured against a baseline. Every stage ends in a decision.

Start with the work

Four questions, four yeses

  • How often does it happen?
  • How much judgement does it take?
  • Can the result be measured?
  • Is the team ready to change it?

Marginal gain — the same work, faster

Structural gain — the unit of work changes

We start with the team that owns the workflow and ask four questions: how often it happens, how much judgement it takes, whether the result can be measured, and whether the team is ready to change how it is done. We also say up front whether the gain will be marginal or structural.

Measure before you build

Baseline sheet, before day one

Time per job
—
Hands it passes through
—
Where it waits
—
Cost per job
—

We record how the workflow runs today: how long it takes, who touches it, where it waits, what it costs. That baseline is the number the pilot is judged against.

Scope a pilot you can finish

Definition of done, day 0Decision
  1. Week 1
  2. Week 2
  3. Week 3
  4. Week 4

Two to four weeks. One workflow, one cohort.

Two to four weeks. One workflow. A named cohort who use it every day. A written definition of done, agreed before day one.

Build inside your systems

In your own account, on

Claude API
Amazon Web Services
Google Cloud

The agent is connected through MCP to the design tools and records the workflow lives in, so one tool’s output becomes the next one’s input. Where a tool has no connector, we build one. It runs in your own account on the Claude API, Amazon Bedrock or Google Cloud Vertex AI. Controls go in on the first day.

Decide on evidence

Baseline

before day one

Result

end of pilot

  • Extend to the next workflow
  • Hand over to your team
  • Stop

Baseline and result, side by side, in your numbers. Then one of three decisions: extend, hand over, or stop. Stopping is a legitimate outcome, and the scope says so.

Hand over so it holds

  1. Supervised
  2. Reviewed
  3. Delegated
Named championsDocumentation the agent reads

Champions named inside your team, documentation treated as part of the deployment, and a staged move from supervised use to delegated work.

Scale when it earns it

An enterprise rollout turns on five decisions, taken in this order, each with a named owner and a start date.

  1. Structure and identity

    One organisation or many; the groups inside it.

  2. Access

    Which surfaces and connectors each group gets.

  3. Governance

    How customisations are controlled.

  4. Spend

    Caps, and who manages them.

  5. Visibility

    What is measured; the signals that show adoption.

  6. Go-live

    Group by group.

Settled first, because they are hard to undo

Domain claimingNumber of organisationsGroup structureData retention

Every open decision is written down with what blocks it, what unblocks it and a decide-by date. The result is a rollout plan your team owns. It also says which answers carry over when a new Claude product arrives.

Then three phases.

  1. Activation
  2. Acceleration
  3. Expansion

Training

Training comes with every implementation

Each role is trained on its own work, at the point in the engagement where it needs it.

  1. During the pilot

    The people who use it

    Working with the agent on their own workflow: what to hand it, how to review its work, when to correct it.

  2. At handover

    The champions who own it

    Keeping the instructions and documentation current, and training the colleagues who come after them.

  3. Before rollout

    Engineering and IT

    Claude Code for building and maintaining connectors, and the admin controls: access, spend and audit.

Controls

A person approves before anything reaches your customer

An audit trail of every agent action, across the whole path

Spend

Spend is capped before it is spent

We set the limits with you during setup, at three levels, and review them with you every month.

What moves spend

  • The default model and effort level
  • Organisation instructions that guide use
  • Reusable skills, loaded only when a task needs them
  • Agentic tasks left running

Sized to real use

Each group's limit fits its typical member. Where usage is unknown we start tight and raise on evidence, with planned room for spiky work such as month-end.

Reviewed every month, in both directions

Every request to raise a limit is questioned, and so is every repeat. Limits that are never reached come down.

Where it runs

Deployed where your data already lives

Claude runs directly on the Claude API, which carries every feature, or inside your own cloud account on Amazon Bedrock or Google Cloud Vertex AI. We set up the path your data and spend already sit in, and build the rest around what it carries.

Claude in your own cloud
Runs inAmazon BedrockYour own AWS accountGoogle Cloud Vertex AIYour own Google Cloud project
Sign-inIAM roles, or bearer tokens for teams without IAM accessGoogle Cloud credentials and IAM
Data residencyA global endpoint, or US, EU, Japan and Asia-Pacific regionsA global endpoint, US or EU multi-region, or a single region
LoggingInvocation logs stay in your account; logging gives neither AWS nor Anthropic access to your contentRequest logs stay in your project; logging gives neither Google nor Anthropic access to your content
CarriesMessages, tool use, caching, thinking, citations, structured outputs. Server-side tools and the MCP connector we run on your sideThe same, plus web search. Code execution and the MCP connector we run on your side
Choose it whenYour data and spend already live in AWSYour data and spend already live in Google Cloud

Get started

Have a workflow in mind?

Tell us what it is, and we will scope a fixed pilot around it.

Discuss a pilot