# Forward Deployed AI & Automation (FDE)

My approach to building AI and automation around the actual workflow, from problem definition to tool integration and verification.

- Canonical: https://www.mehmeterendereli.com/en/services/forward-deployed-ai-engineering
- Language: en

Illustrative technical visual.

Forward Deployed Engineering · AI · Integration

My approach to building AI and automation around the actual workflow, from problem definition to tool integration and verification.

## Approach

Forward Deployed Engineering (FDE) brings technical development close to the people doing the work. I start by examining the process, data sources and bottleneck with its users, then integrate the solution with existing tools. My experience in manufacturing preparation, machining and assembly informs this approach. FDE describes how I deliver AI and automation work.

## Deliverables

- Process map, data and access requirements
- An AI agent, RAG or MCP/API prototype where appropriate
- Failure scenarios, test records and approval points
- Staged rollout, usage and handover notes

## Expected Outcomes

- A defined workflow with a clear problem and scope
- Integrations with inspectable sources and execution results
- Explicit acceptance criteria for success, failure and human review

## How does FDE work for AI automation?

Start with a bounded task such as handling a request, preparing a document or connecting technical tools. Record the manual steps and a baseline. Add RAG for document-grounded answers, MCP/API tool access and human approval where needed. Test missing data, timeouts and invalid outputs alongside the normal path. After user acceptance, plan a staged rollout and measure against the same criteria. Reliable task completion matters more than the choice of model.

## VORMETRA Slice: software interfaces for manufacturing tools

This public OrcaSlicer-based workspace contains G1000 machine and pellet-material profiles plus the vera-control Python layer. Documented HTTP endpoints and MCP tools expose profile lookup, model validation and slicing to other tools. Job locks, timeouts and error handling are inspectable in source. Size validation is a separate step, not automatically required by every slicing call. This is a software example of the integration approach; it does not establish physical G1000 commissioning or measured production performance.

[VORMETRA Slice source](<https://github.com/mehmeterendereli/vormetra-slice>)

[HTTP and MCP interface documentation](<https://github.com/mehmeterendereli/vormetra-slice/blob/main/vera-control/README.md>)

## OpenRelax: desktop automation with explicit boundaries

My Windows maintenance utility uses PowerShell and Windows Forms. Its source exposes the cleanup scope, administrator-permission checks and read-only -SelfTest mode. User approval and file exclusions form part of the workflow. It demonstrates how I design automation with explicit limits and inspectable behavior. Source distribution is available; no measured speed improvement or customer savings are claimed.

[OpenRelax source and usage notes](<https://github.com/mehmeterendereli/openrelax>)

## MED Education: making AI concepts accessible

MED Education is a public application where I turn AI concepts into an interactive learning experience. It offers an example of my work on user-facing AI products. An education application and an industrial integration have different scopes; this link is not evidence of a customer deployment or measured learning outcomes.

[Explore MED Education](<https://egitim.mehmeterendereli.com/>)

## Which projects are a good fit?

Repetitive document work, connecting technical tools, source-grounded access to internal knowledge and operations requiring approval are useful starting points. Choose one process first, then define existing systems, data permissions and acceptance criteria. My manufacturing, PLC/HMI and mechanical refurbishment work continues within its own scope.

[AI agent, RAG and MCP scope](<https://www.mehmeterendereli.com/en/services/ai-agent-workflow-automation>)

[Open-source project evidence](<https://www.mehmeterendereli.com/en/open-source>)

## Can we work remotely, and what comes first?

Discovery, prototyping and review for software and AI integrations can be remote when access conditions allow. Physical-machine intervention and site acceptance require separate planning. In the first discussion we clarify the task, existing tools, an example input and the expected output. Scope, data access and acceptance criteria define the initial piece of work.

## Next Step

### If you have a similar requirement, let's define the scope together.

I review the request and come back with a proposed approach and scope.

[Get in Touch](<https://www.mehmeterendereli.com/en/contact>)

[See All](<https://www.mehmeterendereli.com/en/services>)

## Sources

- [VORMETRA Slice source](https://github.com/mehmeterendereli/vormetra-slice)
- [HTTP and MCP interface documentation](https://github.com/mehmeterendereli/vormetra-slice/blob/main/vera-control/README.md)
- [OpenRelax source and usage notes](https://github.com/mehmeterendereli/openrelax)
- [Explore MED Education](https://egitim.mehmeterendereli.com/)
- [AI agent, RAG and MCP scope](https://www.mehmeterendereli.com/en/services/ai-agent-workflow-automation)
- [Open-source project evidence](https://www.mehmeterendereli.com/en/open-source)