The real work sits between your engineering systems.

I connect requirements, tests, product data and simulation into one traceable digital thread: standards-based, fast to deploy, and documented. And I add AI that runs on your own servers, so your data never leaves your network.

  • DOORS Next
  • ETM
  • PLM
  • OSLC
  • ReqIF
  • Local LLM
  • Docker

Available for projects from December 2026 · Dresden, remote and on-site in DACH and the EU.

Four engineering systems and the gap between themRequirements, test, product data and simulation data sit in four separate systems. At first a person carries data between them by hand: export, e-mail, copy and paste, re-import. Then one traceable thread connects the four systems directly through OSLC and REST, linking requirement REQ-1042 to test TC-211, simulation SIM-088 and part PRT-4711. your network · on-premise by hand export.xlsxe-mailcopy & pastere-import OSLCOSLCRESTOSLC Local AI agent proposes · you decide REQ-1042 v4 → TC-211 suspect signed off by a person · logged Requirements DOORS Next REQ-1042 · v3REQ-1043 · v1REQ-1047 · v2Test ETM TC-211 · passedTC-212 · passedTC-219 · not runProduct data PLM PRT-4711 · rev BPRT-4712 · rev ABOM-88 · releasedSimulation SPDM SIM-088 · convergedSIM-091 · queuedSIM-093 · outdated Four engineering systems and the gap between themRequirements, test, product data and simulation data sit in four separate systems. At first a person carries data between them by hand: export, e-mail, copy and paste, re-import. Then one traceable thread connects the four systems directly through OSLC and REST, linking requirement REQ-1042 to test TC-211, simulation SIM-088 and part PRT-4711. your network · on-premise by hand Local AI agent you decide TC-211 suspect signed off · logged Requirements DOORS NextTest ETMProduct data PLMSimulation SPDM
Before: a person carries data between four systems by hand. After: one traceable thread, watched by a local AI agent whose proposals a person signs off. Point at a system to trace its links.

Integration, compliance and local AI from a single specialist

Engineering toolchain integration

Requirements, test, PLM and simulation data connected through open standards. Nine years of production experience.

  • DOORS Next
  • ETM
  • Teamcenter
  • Aras
  • Minerva
  • OSLC
  • ReqIF
Systems and process

Software for the 2027 deadlines

E-invoicing, Cyber Resilience Act, Data Act and battery passport: fixed scope, typical delivery in 2–10 weeks.

  • XRechnung
  • ZUGFeRD
  • SBOM
  • Data Act API
  • Product passport
Compliance offers

Local AI on your infrastructure

Open-weight language models, retrieval over your documents and agents that propose while a person decides. No data leaves your network.

  • Ollama
  • vLLM
  • RAG
  • pgvector
  • Docker
How local AI works

Where engineering toolchains break

The failures are rarely dramatic. They accumulate, and each one makes the next more likely.

  1. weak links

    Connections between these systems exist, but they are rarely good.

  2. scale

    Performance collapses with large data volumes; old technology hides behind a newer interface.

  3. hand-made

    Every customer configuration is hand-made and must be re-checked at every update.

  4. silent drift

    A change to one requirement silently invalidates tests and simulations that nobody re-runs.

  5. slow escalation

    When something breaks, the fix takes months because the escalation path is long.

I have spent nine years on the other side of these failures: building the connectors, and diagnosing them in live customer environments when a release was blocked.

What I integrate

A map of the systems and standards I work with, as text rather than logos. What matters in PLM integration is which interfaces, object types and versions are involved, and how links behave when a baseline changes.

Requirements

  • IBM DOORS
  • IBM DOORS Next (ELM)

Test

  • IBM Engineering Test Management (ETM)

PLM

  • Aras Innovator
  • Siemens Teamcenter
  • Oracle Cloud PLM
  • PTC Windchill (Covered together with your administrators in the assessment phase)

Simulation data

  • Ansys Minerva

ALM

  • Codebeamer (Covered together with your administrators in the assessment phase)
  • Polarion (Covered together with your administrators in the assessment phase)

Standards and technology

  • OSLC
  • ReqIF
  • REST
  • .NET / C#
  • Java
Unmarked
Nine years of production experience
Marked with a dagger
Covered together with your administrators in the assessment phase

Typical work: DOORS Next integration with ETM and PLM, OSLC integration across vendor boundaries, ReqIF exchange with suppliers, an Ansys Minerva integration that feeds results back to requirements, and Aras Innovator connectors.

Three ways to work with me

Assessment

Fixed scope · 3–5 days

Recommended first step

System landscape, data models, processes, a mapping concept and a reliable effort estimate. You receive a written result that any implementer can work from, including me.

Connector implementation

Time and materials or fixed price

Installation, configuration, attribute and object mapping, workflows, permissions, testing, go-live and hypercare.

Specialist support for system houses

Subcontract · your project lead

Integration capacity under your project lead and, if you wish, under your branding. Second- and third-level support. Feasibility checks and effort estimates for pre-sales.

Day rates and fixed-price assessments on request.

How an engagement runs

Five phases in a fixed order. The bars show the typical range of each phase in working days.

  1. AssessmentLandscape, data models, mapping concept, estimate.
    3–5 working days
  2. Installation & base configurationConnector installed in your test environment; servers, accounts and permissions configured.
    3–5 working days
  3. Mapping & customisationAttribute, object and link mapping and workflows, checked against your real data.
    5–15 working days
  4. Test & acceptanceTests against agreed acceptance criteria, including volume tests on production-sized data.
    4–8 working days
  5. Go-live & hypercareCut-over, monitoring, fixes and handover documentation.
    3–5 working days

Typical total: 18–38 days

The deadlines driving 2027 budgets

Small and mid-sized firms buy software when a law puts a date on it. These are the dates; each has a fixed-scope offer behind it.

  1. EU AI Act

    AI-literacy duties apply to every company that uses AI.

    Local AI pilot
  2. Cyber Resilience Act · Data Act

    CRA reporting of exploited vulnerabilities starts; new connected products must give users access to their data.

    CRA readiness · Data Act connector
  3. E-invoicing

    Companies above €800,000 turnover must issue XRechnung or ZUGFeRD invoices.

    E-invoicing
  4. Battery passport

    EV, LMT and industrial batteries above 2 kWh need a digital battery passport.

    Product passport
  5. Cyber Resilience Act

    All CRA obligations apply to products with digital elements.

    CRA readiness
  6. E-invoicing

    Every German business must issue e-invoices.

    E-invoicing

Saxon SMEs can apply for SAB digitalisation grants covering part of the cost; I help prepare the application.

All offers with scope and timelines

Local AI: your data never leaves your network

Current open-weight language models are good enough for real engineering and back-office work, and they run on a single server you own. I set them up so the assistant can read your documents and systems, while nothing goes out.

Local AI architecture inside your networkInside the boundary of your network: your systems feed a retrieval index; a local language model and an assistant use it; a person approves every proposal and an audit log records it. The connection to external cloud AI services is cut and not required. your network Cloud AI not required Your systems DOORS Next · PLM · ERP Retrieval index PostgreSQL + pgvector Local AI engine Ollama / vLLM Assistant proposes Person decides Audit log records Local AI architecture inside your networkInside the boundary of your network: your systems feed a retrieval index; a local language model and an assistant use it; a person approves every proposal and an audit log records it. The connection to external cloud AI services is cut and not required. your network Your systems DOORS Next · PLM · ERP Retrieval index PostgreSQL + pgvector Local AI engine Ollama / vLLM Assistant proposes Person decides Audit log records
On your hardware
Model, search index and logs run on your server, deployed as Docker containers. Outbound traffic can be blocked completely.
Grounded in your documents
Answers come from your specifications, manuals and tickets, with the source cited, and respect existing access rights.
A person decides
Agents propose changes; a person approves; the system writes; the log records. Nothing is written back without sign-off.
Minimal footprint
One Compose file, one command to install, repeatable updates. A pilot runs in 4–6 weeks.
docker-compose.ymlIllustrative deployment: the whole assistant in one Compose file, with no route to the internet.
services:
  llm:
    image: ollama/ollama
    networks: [internal]
  index:
    image: pgvector/pgvector:pg17
    networks: [internal]
  assistant:
    image: registry.local/assistant:1.0
    environment:
      LLM_URL: http://llm:11434
      SIGN_OFF: required
    ports: ["8080:8080"]  # reachable from your LAN
    networks: [internal, lan]
networks:
  internal:
    internal: true   # model and index: no route out
  lan: {}

Local AI pilot: scope and timeline

Why it is fast: specification first, AI-assisted, signed off by a person

I work from a machine-readable specification. AI coding agents draft implementation, tests and documentation against it; I review, test and sign off architecture and correctness myself. That shortens delivery from months to weeks without handing judgement to a model.

Everything ships as containers with one Compose file and a CI pipeline, so installation is a single command and every update is repeatable. Your project data is not sent to external AI services unless you agree to it in writing.

  1. Specifymachine-readable spec
  2. GenerateAI agents draft code, tests, docs
  3. ReviewI sign off architecture and correctness
  4. Shipcontainers + CI, one command

Synapse: traceability that keeps up with change

In development

Synapse is my own product. It is built for the part of coordination work that connectors cannot do.

Permanent links
between requirements, test cases and simulation results, across tools and baselines.
Change impact
assessed as soon as a requirement changes: which tests and simulations are now suspect.
Coverage gaps
named explicitly: requirements without a test, tests without a current result.
Results returned
from simulation to the requirement they verify, so evidence sits where auditors look.

Every AI suggestion goes through a person

  1. AIproposes
  2. Screenshows
  3. Persondecides
  4. Systemwrites
  5. Logrecords

The language model runs on your infrastructure. No requirement, test or simulation text leaves your company.

Specification complete, implementation under way, pilot customers from Q2 2027.

Interested in a pilot?

The business case (modelled)

For a reference programme of 40 people working with DOORS Next, ETM and a PLM/SPDM system, I estimate about 2,076 hours of coordination work per year.

Transport · ≈ 644 h

Moving data between tools. Today's connectors already handle this.

Judgement · ≈ 1,432 h

Judgement, drafting, searching and providing evidence. This is where Synapse works.

Modelled estimate for a defined reference programme. Assumptions available on request; they will be replaced by measured values at the first pilot customer.

Industries and evidence regimes

The industries differ; the question an auditor asks is always the same: show me, for this requirement, what verified it and in which baseline.

AutomotiveISO 26262 · ASPICE
Functional safety and process assessments both demand requirements-to-test traceability across baselines. I build the links that answer the assessor's question without a spreadsheet.
AerospaceDO-178C
Bidirectional traceability from high-level requirements to verification results, kept consistent across baselines and through ReqIF exchange with suppliers.
Medical devicesIEC 62304
Software requirements, risk controls and verification evidence linked so that the design history file becomes a query rather than a project.
Machinery and plant engineeringEU Cyber Resilience Act
Products with digital elements need documented security requirements and vulnerability handling. Traceability from requirement to test carries that evidence.
SemiconductorsSupplier exchange
Large requirement sets, many suppliers and frequent specification revisions: the workload where ReqIF exchange and change-impact analysis pay off first.
RailEN 50128
Software safety evidence for railway applications follows the same pattern: requirement, design and test, linked and baselined.

Saxony first

Projects run across Germany and the EU, remotely and on site. In Saxony, I am a day trip away from:

  • DresdenSemiconductor fabs and their suppliers, research institutesRequirements traceability for embedded teams; CRA and SBOM for equipment makers
  • Chemnitz · ZwickauMachine builders and automotive suppliersData Act connectors; ASPICE and ISO 26262 traceability
  • LeipzigBattery, e-mobility and automotive plantsBattery passport; requirements work for new programmes
  • Bautzen · GörlitzRail vehicles, production and new research centresEN 50128 traceability; data pipelines and DevOps work packages
Portrait of Dr.-Ing. Waheed Aslam Ghumman

About me

I'm Waheed Ghumman, an integration architect based in Dresden. For nine years I worked as Principal Software Engineer and Solutions Architect for enterprise PLM/ALM integrations, from requirements analysis to production support, for customers in automotive, aerospace and semiconductors.

My work sits where the hard part is not the API call but semantics, versioning and failure behaviour. Root-cause analysis under pressure is what I am most often called in for. From December 2026 I work as an independent consultant.

Facts

Education
Dr.-Ing. (TU Dresden) · M.Sc. Computational Logic (TU Dresden / UPM Madrid) · research at EPFL Lausanne
Experience
Nine years as Principal Software Engineer and Solutions Architect for enterprise PLM/ALM integrations, for customers in automotive, aerospace and semiconductors
Track record
14 years building enterprise integration software. Integrations owned end to end, from requirements to on-site production support. Senior technical authority in architecture workshops and proofs of concept for international enterprise customers.

What I hold to

  • The human decides.
  • The data stays with the customer.
  • Statements about product maturity are honest.

Questions

Do you work remotely?

Yes. Most of the work runs remotely. I come on site in DACH and the EU when workshops, cut-overs or your security rules call for it.

Can you work under our branding as a subcontractor?

Yes. I work under your project lead and, if you wish, under your name towards the end customer. Confidentiality agreements are a matter of course.

How do you handle data from regulated projects?

In line with the GDPR; a data processing agreement is available on request. Data stays in your systems: I work in your environment or with test data. Export-control requirements are respected.

What does an assessment deliver?

A written report covering your system landscape and data models, the processes that touch them, a mapping concept, the risks, and a reliable effort estimate for implementation. It takes 3–5 days at a fixed price and stands on its own, whoever implements afterwards.

Is Synapse available?

Not yet. The specification is complete and implementation is under way; pilot customers from Q2 2027. If you would like to be one of them, I would be glad to talk.

How fast can you deliver?

Compliance offers typically take 2–10 weeks after a one-day deadline check, a local AI pilot 4–6 weeks, and an engineering integration 18–38 working days. Speed comes from a fixed scope, specification-driven work with AI coding agents, and containerised deployment.

Can AI really run without sending our data to the cloud?

Yes. Current open-weight models run on a server inside your network. The model, the search index and the logs stay on your hardware, and outbound traffic can be blocked entirely.

Is funding available in Saxony?

Small and mid-sized companies in Saxony can apply for SAB digitalisation grants for external services and software. The application must be filed before the project starts; I help prepare it, and SAB decides on eligibility.

Let's talk about your toolchain

Tell me which systems are involved, which versions, and where it hurts.

First a call, then a fixed-scope assessment, before anyone commits to implementation.