Architecture review
We review the architecture, code and operations of an existing platform. You get a list of findings and actions, sorted by risk.
[ Senva GmbH · Berlin · since 2015 ]
Senva runs its own AI lab and has built software for companies such as Deutsche Telekom since 2015. Architecture, core development and project management are based in Berlin.
30 minutes with Kim Pham or Holger Spickmann. We reply within one working day.
Services
A self-service portal needs a permission model. An AI assistant needs clean interfaces. That is why our portal, security and AI people work on the same project instead of sitting in separate departments.
Self-service portals, customer portals and internal business applications, connected to the systems you already run. We take care of the architecture, the build and the handover to your operations team.
Proof: trade-fair platform for a telecommunications company, React · Node.js · AWS, 2020 to 2023Portals for digital certificates and smartcards, role and permission models, security reviews of architecture and code. We plan security in before the first line of code, because retrofitting it costs more.
Proof: a self-service portal where end users order and manage certificates and smartcards themselvesComputer vision, machine learning on your data and language models inside existing processes. Language models we recommend to you have been tested in our own lab first.
Proof: visual drone detection pilot for Deutsche Telekom's drone shield programme, 2017After go-live we keep looking after what we built. Service hours and response times are set out in the maintenance contract.
Case study
End users order, configure and manage their security products on their own, without going through support. Senva's team on the portal has seven people: four developers, one UI/UX designer, one project manager and one QA engineer.
How we work
Zone 1
Zone 2
Hosting per project in an EU region, for example on AWS or Azure, or in your own environment. Personal data from your project stays in the EU.
Senva Labs
Whether a language model sticks to rules is not something you will find on a data sheet. You find out by running the experiment. So we do our own research, publish our corrections and, where possible, the code, and use the methods in client projects too.

We recreated Calhoun's mouse experiment with LLM agents and are testing how much of the observed collapse was already written into the prompt.

A language model is trained with reinforcement learning to find vulnerabilities in OWASP Juice Shop, a deliberately insecure test application. We record where this works and where it doesn't.

We test whether a language model keeps to the rules of a tightly restricted artificial language and recognises when it is better to say nothing at all.
Client feedback
Working together was excellent! The team's expertise and creativity took our project to a new level.
From the first idea to delivery, everything was professional and straightforward. Highly recommended!
Their solutions have digitally transformed our company. The team knows how to turn our vision into reality, exactly as we imagined it.
Translated from German.
First steps
Before you award a project, you can test us on a limited task. Each of the three offers has defined deliverables. We agree scope, duration and a fixed price in writing before we start.
We review the architecture, code and operations of an existing platform. You get a list of findings and actions, sorted by risk.
We assess a specific AI use case for feasibility and data readiness, and classify it under the EU AI Act.
An idea becomes a clickable prototype, an architecture sketch and an effort estimate with every assumption stated.
Intro call
Kim Pham runs Senva and has worked on backend systems for more than ten years. In the intro call you work out with him whether and how we can help. If we're not the right fit, we'll tell you in the same call.
We reply within one working day. On request, we sign a non-disclosure agreement before the call.