AI assistants on company-owned infrastructure
I design and integrate fully self-hosted AI solutions with local language models, internal documents, voice functions and tool integrations. The following technical components describe an implemented setup; the selection and operating model are tailored to your business.
The objective was an integrated environment for document work, information retrieval and digital tasks using self-hosted models. Model processing, document knowledge and tools needed a comprehensible interface and controlled data flows.
The environment was developed incrementally, with interface and response-profile adaptations tested on concrete use cases. It combines locally operated language models, document search, dictation and speech output, scheduled tasks and browser tools with manual takeover.
Chat, document knowledge and tools share one environment. Model, document and speech processing run without external model APIs. Browser tools can access external websites; this implementation is not fully internet-isolated. Capabilities, quality and hardware requirements are assessed per deployment.
Automating daily order exports
- Before implementation
- Orders had to reach a downstream process regularly in a prescribed format.
- Challenge
- Time windows, business filters and reliable error handling were central.
- Approach
- Clarify rules, create reproducible test cases, control scheduling and log execution.
- Technical solution
- An automated export with validation, status checks and a clear failure path.
- After implementation
- The recurring process was technically safeguarded; status checks and error messages support ongoing operation.
- Skills
- Shell scripting, SQL, interfaces and operations
Migrating a five-digit customer database
- Before implementation
- A legacy customer database had to move to a new web platform.
- Challenge
- Different structures, incomplete fields and business mappings required controlled rules.
- Approach
- Analyse source data, map fields, validate and perform logged test runs before agreed migration.
- Technical solution
- A repeatable import with plausibility checks and traceable errors.
- After implementation
- Data could be transferred in a structured way and the procedure documented for future runs.
- Skills
- Data processing, SQL, automation and documentation
Modernising a specialised online shop
- Before implementation
- An established shop needed technical renewal while preserving its business-specific workflows.
- Challenge
- Historical data, customer-group processes and interfaces had to be considered together.
- Approach
- Inventory the system, plan a phased migration and repeat tests before handover.
- Technical solution
- A WordPress/WooCommerce platform with custom imports, role logic and documented operations.
- After implementation
- The platform was modernised and individual workflows and operational knowledge documented.
- Skills
- Web development, migration, e-commerce and quality assurance
Taking over an established web platform
- Before implementation
- A live website needed reliable technical support after several development phases.
- Challenge
- Dependencies, hosting and custom extensions were only partially documented.
- Approach
- Back up the existing state, assess risks, prioritise fixes and build documentation.
- Technical solution
- A controlled update process, version control, backup concept and gradual remediation.
- After implementation
- Changes became easier to control and the technical state clearer for future maintenance.
- Skills
- WordPress, hosting, security and maintenance
Developing a nutrition press service
- Before implementation
- A nutrition press service combines industry monitoring with journalistic selection and editing.
- Challenge
- Capture automated news sources reliably, detect duplicates and preserve provenance in an editorial workflow.
- Approach
- Record sources and usage rights, automate feed processing gradually and retain editorial approval.
- Technical solution
- News feeds with normalisation, topic assignment, duplicate detection and tools for review, editing, approval and corrections.
- After implementation
- Recurring collection work is reduced while selection, interpretation and publication remain editorial responsibilities.
- Skills
- Feed automation, data processing, editorial workflows, web development and source management
Investigating an ASLR weakness in the Apple kernel
Independent security research examined address-space randomisation in the kernel of several Apple operating systems. The work identified an ASLR weakness as part of system-level analysis of memory-protection mechanisms.
Apple acknowledges Joshua Baums of Informatik Baums in its kernel acknowledgements for iOS 15 and iPadOS 15, watchOS 8, tvOS 15 and macOS Monterey 12.0.1.