Leon Drong

ERP, integrations and AI, from operations to delivery

I connect applications, data and processes, above all in ERP, integrations, Atlassian and AI.

leon@drong.dev

Selected work

  1. Lucas-Nülle GmbH, since 2025

    NAV project data for time tracking

    in production

    Technology

    • C#
    • REST
    • OAuth 2.0
    • SQL
    • Dynamics NAV

    Outcome

    New projects are ready for time bookings in the HR system right away. Filtered records and error messages go straight to the responsible team.

    Context, contribution and decision
    Context
    The previous HR system was cancelled and replaced by a new one. There was no transfer of project data from Dynamics NAV into the new system.
    My contribution
    I built the connection from Dynamics NAV to the HR system, including data mapping, SQL queries, error handling and monitoring.
    Decision
    Data moves over REST with OAuth 2.0. All projects are uploaded first, then the status is checked with a delay because of the API limits.
    Process in 6 steps
    New projects move from Dynamics NAV into the HR system after checks. Filtered records and error messages go to the responsible team.
    1. Project created in Dynamics NAV

    2. Projects collected via SQL

      A SQL query finds new projects.

    3. Data checked and cleaned

      Faulty data is transformed or filtered out.

      Records filtered out

      The responsible team is notified.

    4. Uploaded one by one

      Through the HR system REST interface with OAuth 2.0.

    5. Status queried after a delay

      Only after all uploads, because of the API limits.

    6. Status evaluated

      Every status other than waiting is logged.

      Error reported

      The responsible team is notified.

  2. Lucas-Nülle GmbH, since 2024

    Scanner application for manufacturing

    in pilot

    Technology

    • C#
    • Dynamics NAV
    • SQL

    Outcome

    Bookings are made directly from the workstation into Dynamics NAV, and the storage system takes the data over from there. The application has been tested in production.

    Context, contribution and decision
    Context
    Bookings depended on a computer at the edge of the production floor. Even a single transaction therefore involved an unnecessary walk.
    My contribution
    I developed and tested the application and its integration with the ERP system.
    Decision
    The interface is designed around the scan at the workstation. Validation and feedback happen immediately before the data is passed to Dynamics NAV.
    Process in 4 steps
    Simplified flow from scan to storage system, details are omitted out of respect for the company.
    1. Transaction scanned at the workstation

    2. Check and upload with filters

      The app checks the data before it is booked.

      Filter flags the data

      Error message asking for a correction.

    3. Booking in Dynamics NAV

      The app talks directly to the ERP.

    4. Storage system takes over the data

      The storage system pulls the data from Dynamics NAV.

  3. Lucas-Nülle GmbH, since 2024

    AI governance with inventory and approval workflow

    ongoing

    Technology

    • AI governance
    • EU AI Act
    • Zero Trust
    • Jira

    Outcome

    Every AI system in use has a documented classification with conditions. Requests and decisions are kept in one place. A few tools are set as the standard, so not every request leads to another product. It does not replace a legal review.

    Context, contribution and decision
    Context
    There was no overview of which AI is used for what, how it was assessed and under which conditions it is approved.
    My contribution
    I built the assessment framework and approval path and maintain the AI register with assessed systems. I set conditions and make recommendations. Management decides. I also create knowledge pages and training content on AI literacy.
    Decision
    The framework combines the EU AI Act with an internal risk class and zero trust minimum measures. It builds on the data classification and is documented in Jira.
    Process in 6 steps
    Every AI request is assessed, documented and decided by management.
    1. AI application requested

      The requester names the concrete use case.

    2. Assessed by zero trust and AI Act

      Based on the data classification.

    3. Checked against the AI register

      Not yet documented

      Register entry with measures based on risk.

    4. Recommendation with conditions

      I recommend approval or rejection.

      High risk level

      The works council is involved.

    5. Decision by management

      Also for documented cases, because it owns the budget.

      Rejected

      The application is not approved.

    6. Use after approval

      The recorded measures apply on activation.

Further work

  • Lucas-Nülle GmbH, since 2024

    IT service desk with Jira Service Management

    ongoing

    I am rebuilding the IT service desk close to the product standard for two companies and decide on structure, workflows and approval paths.

    The rebuild is still in progress. The goal is a much more efficient handling of requests.

    Details

    Every request becomes a ticket with its own form per request type, an urgency and deadlines in core hours. The Atlassian AI agent Rovo assigns uniform titles, summarises tickets and lists open questions and next steps. Solving low risk requests automatically is planned. The portal is available in German and English.

    Technology

    • Jira Service Management
    • Jira
    • Confluence
    • Automations
    Process in 7 steps
    Every request becomes a ticket, Rovo prepares it, and the deadline only runs in core hours.
    1. Request to IT

      Through the portal in German or English.

    2. Ticket in the portal

      Every request becomes a ticket, with its own form per request type.

    3. Urgency set

      High, Medium or Low. High means work is not possible.

    4. Rovo prepares the ticket

      Uniform title, summary, open questions, next steps.

    5. Deadline in core hours

      The deadline depends on the urgency, Low is the default.

    6. Manual assignment

      Internal comments only for IT and those involved, few notifications.

    7. Low risk requests solved directly

      planned

      Rovo is meant to solve low risk requests on its own.

  • Lucas-Nülle GmbH, since 2026

    Move from Dynamics NAV to Business Central

    ongoing

    The move is a complete rebuild, away from the grown custom system towards a stable, clean system. Together with a colleague I am responsible for data quality and migration, integrations, AL development and the software architecture design.

    The project is still running. Results will be added here once they are reliable.

    Details

    The move runs as a project with an implementation partner. To prepare the migration I take care of data cleansing and data quality. Add-on and add-in management, process consulting with the departments and test scenarios that I define and run myself are part of it as well. I prepared the structures of the existing NAV database so the partner could assess the current landscape.

    Technology

    • AL
    • Dynamics 365 Business Central
    • SQL
    • Data migration
  • Lucas-Nülle GmbH, 2026

    E-invoicing and data export for tax audits

    completed

    Electronic invoicing and the data export for tax audits under GoBD and GDPdU are in place. I supported the development in a small role.

    Details

    Technology

    • Dynamics NAV
    • XRechnung
    • SQL
    • GoBD
  • Lucas-Nülle GmbH, since 2024

    Local dictation tool

    in pilot

    I built a dictation tool that converts speech to text with Whisper on local hardware.

    The tool is not in operation yet. The first users have been trained on it.

    Details

    A domain glossary keeps NAV and Business Central terms correct. Obvious hallucinations such as words repeated suspiciously often are removed. The text goes into the chosen text field or into the history.

    Technology

    • Python
    • Whisper
    Process in 4 steps
    Speech becomes cleaned text on local hardware.
    1. Speech recorded

    2. Converted to text locally

      Whisper with a glossary for NAV and Business Central.

    3. Hallucinations removed

      For example a word that repeats suspiciously often.

    4. Text into the chosen field

      History chosen

      The text is stored in the history.

  • Personal project, since 2019

    Personal multi-AI system

    ongoing

    I am developing a personal multi-AI system. EVORA plans as the central instance and is meant to delegate to specialised AI experts, VELEN checks the results independently.

    The system is already running and is being extended. Connecting it to everyday life is currently held back by hardware costs.

    Details

    EVORA stands for Evolving Virtual Orchestration and Reasoning Architecture.

    VELEN stands for Verification, Explainability, Logic, Evaluation and Negation.

    The project began in 2019 with scripts and automations. The language-model layer came later and keeps models replaceable. Tools have separate permission levels and critical actions require approval.

    The first priorities are expert selection, structured handovers between agents and permissions per expert. Model selection by cost, a shared knowledge store and the automatic creation of new experts follow later.

    Technology

    • Python
    • Retrieval
    • LLM agents
    • Evaluation
    Process in 4 steps
    EVORA coordinates the experts, VELEN checks the results.
    1. Analysis

      Understand the request, define requirements and success criteria.

    2. Planning and delegation

      planned

      EVORA picks suitable experts and distributes the tasks.

      Skill missing

      Later EVORA is meant to create a new expert and test it in isolation first.

    3. Validation

      VELEN checks independently, errors go back for correction.

      Critical action

      Carried out only after explicit approval.

    4. Integration and improvement

      Merge the results, lessons improve the expert network.

  • alimex, 2024

    Automated gate access prototype

    prototype

    My IHK final project recognised licence plates from a continuous camera feed. I trained my own YOLO model on plates from Germany and its neighbouring countries.

    The project remained a prototype and was not introduced as a production system.

    Details

    I started with TensorFlow and switched to YOLO, because it was the better solution. The recognised string was checked against a list from a role-based interface, and arrival and departure were recorded with timestamps.

    Technology

    • Python
    • OpenCV
    • YOLO
    • Haar Cascades
    • EasyOCR
    Process in 7 steps
    From camera feed to barrier, checked against a maintained list, with timestamps for arrival and departure.
    1. Camera feed read

      Continuous video at the entrance, in greyscale.

    2. Plate detected

      Self-trained YOLO model and Haar cascades.

    3. Characters read with EasyOCR

      Easily confused characters such as W and H, O and 0 or O and Q are checked.

    4. List and permission checked

      List from a role-based interface plus an entry permission.

      Unknown or not permitted

      Barrier stays closed, the plate is logged and listed for approval or rejection.

    5. Barrier opens

      A one-time access is disabled afterwards.

    6. Arrival with timestamp

    7. Departure with timestamp

      A second camera at the exit recognises the vehicle.

  • Client project, part-time, 2022

    LiDAR obstacle detection for drones

    completed

    I built LiDAR obstacle detection and avoidance for medical delivery drones of an international client. If an obstacle gets too close, the detection takes over the existing automatic control. Afterwards the normal control system continues.

    The drone is in use at the client.

    Details

    I built and tested it on my own, from the sensor and the wiring of a test rig to the code. I deliberately wrote the code without a framework, as a learning challenge. The project work with the client was in English.

    Technology

    • Python
    • Raspberry Pi
    • LiDAR
    Process in 4 steps
    The LiDAR detection only steps in when an obstacle gets too close.
    1. Drone flies automatically

      The existing control system steers the drone.

    2. LiDAR measures the distance

      No obstacle too close

      The control system keeps flying unchanged.

    3. Control overridden

      If an obstacle gets too close, the detection takes over.

    4. Normal control continues

  • Sep 2019 to Sep 2024

    Part-time independent technical projects

    completed

    My part-time self-employment included a Discord bot for moderation and workflows with CI/CD as well as the LiDAR obstacle detection for drones. In 2024 the focus was middleware, computer vision with Python and process automation.

    Details

    Technology

    • Python
    • Middleware
    • Computer Vision
    • CI/CD
  • Next Levels, Aug 2021 to Feb 2023

    Technical project work during my apprenticeship

    during the apprenticeship

    During the Next Levels part of my apprenticeship, I worked on technical project planning, estimates, frontend development and content management for the training company’s client projects. This included work on sites and apps for Mellerud, ENRA, Kumho Tire Europe, Bike Discount and Vosdellen.

    These brands were projects of the training company and were not my own clients.

    Details

    Technology

    • Technical project planning
    • JavaScript
    • CSS
    • CMS

Built on the side

  • Home Assistant on HAOS

    My own home automation, for example pantry management with Grocy, ventilation monitoring and automatic recovery for Zigbee devices.

  • Relief amount app

    Tracks the relief amount of the German long-term care insurance, because there is no suitable solution for it yet.

  • Group schedulerin progress

    Finds the best common date from everyone’s availability, without an account. Built for a tabletop role-playing group.

  • FieldBestiaryin progress

    A digital expedition journal for a fictional world with creatures, ecosystems and field notes.

  • Small tools

    Automatic image sorting and system-wide hotkeys for everyday use.

Background

Positions

  1. since Oct 2024

    ERP Application Administrator

    Lucas-Nülle GmbH

    I run Dynamics NAV in production with C/AL and SQL, build integrations and work on the move to Business Central. I built the Atlassian platform and own it as product owner for two companies. AI governance, the service desk and the security of the applications I look after are part of the role as well.

    View related work

  2. Sep 2019 to Sep 2024

    Part-time self-employed

    Technical projects alongside my orientation period and apprenticeship. The focus in 2024 was integrations, process automation and computer vision.

    View related work

  3. Apr 2023 to Jun 2024

    Apprenticeship in application development

    alimex

    Django, JavaScript, computer vision and fault analysis in live operations.

    View final project

Earlier positions
  1. Mar 2023

    Internship before continuing the apprenticeship

    alimex

  2. Aug 2021 to Feb 2023

    First part of my apprenticeship in application development

    Next Levels

    Frontend development, CMS work, technical project planning and estimates in the training company’s client projects.

    View related work

  3. Apr 2021 to Jul 2021

    Preparatory internship

    Next Levels

  4. Sep 2019 to Mar 2021

    Orientation period

    Several internships and a temporary role in warehouse picking. My part-time self-employment ran alongside this period.

Education and training

  • Feb 2026 to Dec 2026, expected

    Advanced training course in Microsoft Dynamics 365 Business Central

    Continuing education programme, ongoing, get&use Academy GmbH, part-time

    Modules and evidence
    • Application and processes

      Business Central, financial management, SCM and further application modules.

      Modules

      • Business Central Essentials

      • Finanzmanagement, Essentials und Advanced

      • Anlagenbuchhaltung

      • SCM, Essentials und Advanced

      • Bedarfsplanung

      • Produktion

      • Logistik

      • Lagerbewertung

      • Service Management

      • Relationship Management

      • Subscription Billing

      • OPplus

    • AL development

      AL Introduction, Solution Development and Reporting.

      Modules

      • AL Introduction, Foundation und Advanced

      • AL Solution Development

      • AL Reporting, Foundation und Advanced

    • Exam preparation

      Preparation for the MB-800 exam. It is not a passed exam and not a Microsoft certification.

      Modules

      • Functional Consultant Zertifizierungsvorbereitung MB-800

      Evidence

      • Functional Consultant Zertifizierungsvorbereitung MB-800

        Exam preparation, Issuer get&use Academy GmbH, Issued

  • Mar 2026

    Jira automations

    Training, completed, Communardo

    Training in the design, triggers, conditions and maintenance of automation rules.

  • Feb 2026

    Certified Database Practitioner

    Passed certification, completed, DB Academy

    Completed certification covering database fundamentals and SQL.

  • Feb 2025

    Data protection fundamentals

    Training, completed

    Foundation training in handling personal data in line with data protection requirements.

  • Oct 2024

    XRechnung

    Training, completed

    Training in the fundamentals and processing of electronic invoices in the XRechnung format.

  • Jun 2024

    IT specialist for application development (Fachinformatiker)

    IHK Mittlerer Niederrhein

    Passed the IHK vocational qualification in June 2024.

  • Sep 2021 to Jun 2024

    Vocational school qualification

    Berufskolleg Platz der Republik für Technik und Medien, Mönchengladbach, Final school grade 2.0. This grade applies to the school qualification, not the IHK exam.

    School-based part of the apprenticeship in application development.

  • Aug 2017 to Sep 2019

    University of applied sciences entrance qualification

    Berufskolleg Viersen Dülken

    Focus on electrical engineering and information technology.

Tools

  • ERP and data

    Dynamics NAV, C/AL, SQL Server, SQL, Access (used in production)

  • Business Central

    Dynamics 365 Business Central, AL, Data migration (training and migration preparation)

  • Development

    C#, Python, PowerShell, JavaScript, HTML, CSS, Django, CMS (used in project work)

  • Integrations

    REST APIs, OAuth 2.0, ETL, Middleware (used in project work)

  • Atlassian

    Jira, Jira Service Management, Confluence, Automations (operations and administration)

  • AI governance

    EU AI Act, Zero Trust, AI register, Risk classification (used in project work)

  • AI applications

    LLM agents, Retrieval, Whisper, Pyannote (used in project work)

  • Computer vision

    OpenCV, YOLO, EasyOCR, TensorFlow (used in project work)

  • Engineering practice

    Git, CI/CD, Logging standards, Technical project planning, Troubleshooting (used in project work)

  • Languages

    English (fluent in project work)

Personal

Outside work I play tabletop role-playing games. Until 2023 I also coached a junior League of Legends team.

Roll the die.

Contact

Write to me if your work involves ERP, integrations or AI. A short email with the topic is enough.