Applied AI Engineering

We engineer AI products, ML systems and platforms.

Cognitrace works on systems where AI has to perform reliably as part of a product or platform—not in isolation.

Technical advisory, architecture and implementation come from one experienced team.

Discuss a project View projects

13+ years building production software · AI/ML since 2017 · former AI R&D at trivago

Selected companies and products from our work
Projects

Selected projects in AI, ML and platform engineering.

Current Cognitrace projects and earlier roles held by the team. For each example, we state our specific contribution.

Earlier role: trivago
trivago Image Concepts

Computer-vision platform for more than 100 million hotel images

100M+
Images processed
90+%
Precision on the defined evaluation set
End-to-end
Training through deployment

Our contribution: AI/ML R&D, training and data pipelines, distributed processing and deployment.

Cognitrace project
360 Health & Performance

Digital health platform with personalised AI guidance

24/7
Personalised guidance
≤ 2 ms
HRV inference within the analysis pipeline
All defined core tasks
passed in the acceptance test

Our contribution: product architecture, AI orchestration, application, inference integration and cloud platform.

Current enterprise project

Enterprise AI platform for multiple models and applications

Our contribution: platform architecture, model routing, evals, observability and cost controls.

Earlier client project
E-mobility platform

Cloud and integration platform for European e-mobility

100k+
Active users
60s → 1–2s
Latency reduction
Across Europe
Roaming and charging-network integration

Our contribution: backend architecture, cloud infrastructure, roaming integrations and performance optimisation.

Discuss a project We share further technical details and references once approved.
Expertise

What we take ownership of

Architecture, technical leadership and implementation across four areas.

01 GenAI Products & Product Capabilities

Generative AI as a real product capability

We build applications and product capabilities using language and multimodal models, retrieval, tool use, structured workflows and evals.

02 ML & Computer Vision

Specialised ML systems from training to production

Models, data and training pipelines, computer vision, ranking, retrieval, high-throughput inference, evaluation and MLOps.

03 AI Platforms & Cloud

Shared technical foundations for multiple teams and products

Models, data, identity, integrations, deployment, evals, monitoring and cost controls in one shared platform architecture.

04 Technical AI Advisory

Feasibility, target architecture and path to implementation

Technical feasibility, build vs buy, architecture decisions, technology selection, reviews, migration paths and transition into delivery.

Discuss a project
When Cognitrace fits

When AI becomes part of a product or platform.

01 A new AI capability needs to become a lasting part of a product or platform.
02 An existing ML or GenAI system is reaching limits in quality, latency, cost or reliability.
03 Multiple models, applications or teams need a shared technical foundation.
04 The current architecture makes integration, scaling or further development difficult.
05 A critical AI initiative needs technical leadership and additional delivery capacity.
06 The internal team should be able to own the code, architecture and operations afterwards.

Typically not: standalone tool rollouts, no-code automation, training programmes or staff augmentation.

References

What clients say about working with us

"Bringing in AI took 360 to a new level. Users stay in control and see what matters early. Mickey and the team listened very carefully and delivered with real expertise. Stellar work."
Justin Buckthorp · Founder & CEO, 360 Health & Performance
"They engaged constructively and quickly understood the problems we actually wanted to solve."
Elio Santana · Technical Lead, Concentrix Tigerspike (Sydney)
Company

Experienced engineers who advise and build.

Cognitrace is a founder-led engineering team for applied AI. We work directly with product and engineering teams on systems that require AI/ML, software and cloud expertise in one team.

Founded by Mickey Graf, AI Systems Engineer and three-time technical founder. More than 13 years building production software, AI/ML since 2017 — including AI R&D at trivago and projects for startups, mid-market companies and international enterprise programmes.

Technical decisions are tested in working software. Code, documentation and knowledge are then transferred to the internal team.

How we work
  • Experienced engineers work hands-on in the system.
  • Architecture is validated through code and measurements.
  • Quality, latency and cost are evaluated systematically.
  • The internal team remains technically independent.
FAQ

Questions about working together

What kinds of projects fit Cognitrace?
Cognitrace fits when a proprietary AI capability, ML system or platform needs to be built or improved — and model quality, software, data, integrations and cloud operations belong together.
Do you only advise, or do you also implement?
Both, with a clear bias towards implementation. Advisory and technical leadership are also available as a standalone assignment when an existing team owns delivery.
Can you take over an existing system or prototype?
Yes. The starting point is a technical review of architecture, quality, operability and a path forward — followed by handover or focused implementation.
How do you work with existing product and engineering teams?
We integrate directly with existing teams rather than building a parallel stack. Responsibility is cut cleanly — for architecture, technical decisions and the areas where we implement.
How do you measure quality, latency and cost?
Through evals, production telemetry and cost attribution as part of the system: test sets, metrics and monitoring on output quality, error behaviour, latency, cost and tool usage.
Who owns the code, infrastructure and documentation?
You do. Code, infrastructure, models and documentation transfer to the internal team and can be operated without Cognitrace.
How do you handle security, data protection and regulatory requirements?
As part of the architecture: through data flows, roles, deletion concepts and audit trails. GDPR and the EU AI Act are classified per use case. Evidence is produced continuously, not after the fact.

More questions

Contact

Are you planning or scaling an AI product, ML system or platform?

Briefly describe the initiative, its current stage and the biggest technical hurdle. We will say openly whether Cognitrace is a fit and what a sensible starting point looks like.

Discuss a project Email us
Email
contact@cognitrace.de Get in touch Reply within 24 hours.