Showroom
Challenges we have solved.
Selected projects, deliberately anonymised: no clients and no industries. What matters is the task, the requirements and the route to the solution.
- Data & analyticsPilot
Controlling cockpit with an AI data analyst
Finance and sales figures were spread across several databases and hard-to-read fact tables. Management needed a consolidated live view and answers to ad-hoc questions without writing SQL.
- Python
- FastAPI
- SQL Server
- LLM-Agent (Text-to-SQL)
- AI & agentsPilot
AI voice agent for inbound calls
Inbound calls were to be answered around the clock in natural language: answer frequent questions, capture call-back numbers and messages, and hand requests to the right people, with nobody sitting at the phone.
- Node.js
- Fastify
- WebSocket
- Telefonie-API
- Industrial & hardwarePrototype
Vision-based inspection with a camera gantry
Stock had to be reconciled with the target state in the system on a regular basis. Counting and scanning by hand was slow and error-prone, especially across large areas.
- Python
- FastAPI
- Kotlin
- CameraX
- Industrial & hardwarePilot
Automated drilling unit with a safety concept
A manual machining step in an existing production setup was to be automated as a machine of its own: repeatable sequences, safe emergency-stop logic and operation that works without training.
- Python
- FastAPI
- OPC UA
- SPS (IEC 61131-3 ST)
- Industrial & hardwarePrototype
Cashless payment at vending machines via QR
Existing machines were to accept cashless payment without expensive proprietary card terminals. Payment happens in the smartphone browser; dispensing on the device has to be triggered tamper-proof.
- TypeScript
- Fastify
- Prisma
- MQTT
- AI & agentsPilot
AI-assisted parametric CAD environment
Existing CAD programs get AI as an add-on. What is missing is a shared, unambiguous language between human, AI assistant and geometry: which face is meant, what does ‘the left hole’ become after the next rebuild? AI suggestions are hard to verify, and changes to individual features are error-prone.
- Python
- CadQuery/OpenCascade
- MCP
- FastAPI
- AI & agentsPrototype
Secure connection of a document management system to AI
Staff were to access documents in the document management system through an AI platform. Strict confidentiality and permission requirements applied, and every technical claim had to be evidenced.
- MCP
- Node.js
- REST-API
- Azure
- Automation & integrationPilot
AI routing of call requests to the right person
An AI phone assistant took requests, but forwarding them to the right person was manual. The goal was automatic, traceable assignment based on the organisational structure.
- Python
- FastAPI
- LLM
- Messenger-API
- Automation & integrationPilot
Lead research with web enrichment and LLM extraction
Sales needed qualified contacts in defined target groups and regions, systematically. Manual research was slow, incomplete and produced duplicates of existing records.
- Python
- Qt/QML
- Places-API
- LLM (optional)
- Automation & integrationPrototype
Agentic CI/CD pipeline with monitoring
Several software projects were to be developed quickly with AI coding agents, yet under control. That required uniform quality gates, deployments and observability.
- GitHub Actions
- Docker
- Grafana
- Prometheus
- Platform & webPilot
Self-hosted intranet with Microsoft 365 integration
Information, documents, dates and contacts were to be central and easy to find. The standard collaboration platform was too heavy, but its data still had to be used.
- Next.js
- TypeScript
- Microsoft Graph
- OIDC/SSO
- Platform & webPrototype
Lean on-premises CRM with Windows sign-in
Customer relationships, contacts and opportunities were to be maintained centrally, without a cloud CRM and linked to the customer numbers in the ERP.
- Node.js
- SQLite
- IIS
- Windows-Authentifizierung
- Automation & integrationPilot
Automated workstation setup
Setting up new or reinstalled Windows machines by hand was slow, inconsistent and poorly documented.
- PowerShell
- WPF
- winget
- Data & analyticsPilot
Social media monitoring with AI analysis
Activity and reach of own and third-party channels were to be captured across platforms and compared. The platforms deliver their data through different, frequently changing APIs.
- Node.js
- Next.js
- MongoDB
- Chart.js
- Data & analyticsPilot
3D terrain analysis with grid planning
Areas had to be measured directly on real 3D terrain and regular grids derived from them. Planning in plan view distorts distances and areas.
- TypeScript
- React
- CesiumJS
- Web Worker
- Platform & webPilot
Marketplace platform for campaigns
Brands and content creators were to plan, run and evaluate campaigns together. Profiles, media, metrics and the shop connection were scattered across many tools.
- Node.js
- Next.js
- MongoDB
- Cloud Storage
- AI & agentsPrototype
Visual AI pipeline for content scripts
Producing short-form video content for several platforms regularly took a lot of time for ideas, scripts and platform-specific variants. The goal was an automatable, white-label solution.
- Next.js
- TypeScript
- React Flow
- Prisma
- Platform & webPrototype
Smartphone camera as an encrypted network stream
A smartphone was to act as a live camera, without an account, cloud service or extra software, viewable in any browser on the local network.
- Kotlin
- Jetpack Compose
- Swift
- SwiftUI
- Data & analyticsIn production
Data access library with vector search
Several applications needed uniform, safe database access including full-text and vector search for AI features, without rewriting that logic in every project.
- Node.js
- Mongoose
- MongoDB Atlas Search
- Vektorsuche
All projects are anonymised. We do not name clients, industries or locations. Reference calls on request, if the client agrees.
Contact
Which workflow is holding your team back?
Describe it in three sentences. Within 24 hours you get an honest assessment of whether and how we can help.