We have found a way to realize multi-level, explicit reasoning in the combination of a stand-alone reasoning engine and existing AI language models of the GPT4 class.
The interaction of AI language models, structured knowledge and a dedicated reasoning engine recreates central functional blocks of the human brain. This is why we call our approach “bionic”. It raises the performance of machine ‘understanding’ and ‘thinking’ to a new level.
Compared to so-called large reasoning models (such as o1), the bionic approach has the advantage that no expensive and time-consuming training is required; instead, the increase in intelligence is realized purely at the software engineering level. This enables shorter and more frequent innovation cycles, improved controllability and significantly higher efficiency.
The interaction of AI language models, structured knowledge and a dedicated reasoning engine recreates central functional blocks of the human brain. This is why we call our approach “bionic”. It raises the performance of machine ‘understanding’ and ‘thinking’ to a new level.
Compared to so-called large reasoning models (such as o1), the bionic approach has the advantage that no expensive and time-consuming training is required; instead, the increase in intelligence is realized purely at the software engineering level. This enables shorter and more frequent innovation cycles, improved controllability and significantly higher efficiency.
Stand-alone reasoning
Embedding knowledge (grounding)
Function Calling (Agentic)
Overview
Stand-alone reasoning
Embedding knowledge (grounding)
Function Calling (Agentic)
Overview
With the AI MicroWorker® Pattern, we present an innovative, hybrid architecture for advanced intelligence. On this basis, our systems are capable of multi-level reasoning, plausibility checks and validations. Embedded function calling enables agent-like automation with the option of explicitly validating steps before they are executed.
The combination of outstanding performance and reliability makes the AI MicroWorker® Pattern a key to bringing advanced intelligence to business-critical processes and regulated environments.
Complex tasks are broken down into individual sub-tasks.
Each sub-task is contextualized with individual knowledge.
The knowledge is interpreted in the context of the specific sub-task.
The individual interpretation is checked for plausibility at the content level.
After a successful plausibility check, a formal validation is carried out against individually configurable behavioral guidelines.
After successful validation, the final conclusion is drawn and the result is passed on to the next sub-task.
The step-by-step preparation and structured storage of refined knowledge is an important prerequisite for reliable AI use: in this way, the general knowledge from the AI language models can be supplemented with relevant facts and individual specialist knowledge.
In the same way, the laws, regulatory requirements and individual company guidelines to be observed for the automation of processes / agentic workflows are stored in a structured manner.
Unstructured information is extracted in a structure- and context-sensitive manner and, if necessary, cleaned up for further processing.
Raw information is gradually enriched into knowledge fragments on the basis of ontology structures.
The enriched knowledge fragments are stored in a “retrieval-friendly” format for AI systems.
The systematic networking of individual knowledge fragments results in high-quality, condensed knowledge.
AI language models are an important but ultimately interchangeable commodity component of the embraceable platform. Thanks to our multi-source AI paradigm, we are ‘model agnostic by design’, giving you maximum flexibility and freedom of choice. The same applies to the underlying compute infrastructures.
In addition, we attach great importance to ensuring that no data is persisted by the model operator or even used for model training. So you remain completely “data sovereign”. More flexibility and freedom of choice is not possible!
The embraceable platform can be used in combination with all leading AI language models and embedding models.
Thanks to the model-agnostic approach, you can also use open-source models such as Llama3.1 450B or Mixtral 7x8B without any problems.
(AI Model Hub)
Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod
Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod
Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod
Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod
Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod
