AiSEC is our artificial intelligence solution, fitted to your company, keeping your data safe.
¶ Why was AiSEC developed?
We, as Sphinx, have made it our mission, to develop innovative solutions which are tailored to our customers. You can find out, how we approach such challenges in "The Sphinx Approach".
One of such solutions is AiSEC. Nearly everyone nowadays uses a chatbot for some use-case or another even for work related tasks and questions. And while artificial intelligence offers a lot of benefits in various fields and can ease the workload, one also has to carefully think about the data one provides to such applications, as this data is collected by them. No one can guarantee that the whole conversation one has with a common chatbot is not saved and used for learning, but what is guaranteed is that the metadata is used that way. Therefore, one has to consider, for each question they pose to a chatbot, if any of the data contains or implies confidential data, data that when falling into the wrong hands, could lead to losing ones advantage over the competitors or data that when made public violates the GDPR. Such an example is discussed in more detail further down in this article.
As these are a lot of aspects, which an employee would have to keep in mind for each question he or she poses or for each task the artificial intelligence is used to help with, we, as Sphinx, have developed AiSEC as a solution to such problems. This privacy concern is just on of the three main reasons AiSEC was developed. With AiSEC your data stays within your company and won't fall into the wrong hands, as this solution is deployed on-premise.
Furthermore, RAG, retrieval-augmented generation, can be implemented, making it possible to grant access to specific company-related data to the model, without having to re-train it. Instead it just expands its knowledge base with the data provided to it, making it possible to generate company specific answers to questions. An example would be to add company data about customers to the model, specifiying which of the data is relevant to distinguish between customer groups and than let the model use this information to generate whitepapers specifically targeting each of the customer groups. Another use-case would be, feeding data about orders and shipping into the model using it to get a notification if one of the inventories is running low and the model has found a discount for this exact product somewhere during its web search.
The last main reason for the development of our solution is video and image generation. While most solutions do offer this in some way, our solution easily transforms text into an image, with just one click. Furthermore, it is based on stable diffusion, therefore offering a multitude of possibilities to tweak and configure the image outcome. With this solution one cannot only tell the artificial intelligence what should be depicted on the image or in the video, but also what should not be in it and one can do so via human readable text, simplifying the handling.
¶ Example
An example for a scenario that can easily happen and propbably is also often happening nowadays in many companies where Large Language Models (LLM) such as ChatGPT are used is the following:
An employee wants to get possible improvements to a flyer he or she is creating, which is used to advertise a new and innovative solution the company has been working on and which it wants to promote. Therefore, the employee provides all the information, that should be included in the flyer to a LLM such as ChatGPT, also providing background information about the product, so the flyer presents as much information as possible. Furthermore, the employee also defines the target group, which should be reached with this flyer. Now, the LLM might create a well-designed flyer fitted to the target group, however the LLM also learns that this specific target group can be reached with this type of innovative technology and therefore will provide this technology and this flyer design as an answer to future requests about this target group from companies belonging to the same sector.
This way not only does a company loose its advantage based on the innovative idea, which is an intellectual property of that company, and which is now part of the knowledgebase of the LLM which was used, meaning this solution will also be presented as a solution to other companies, but also the new flyer design will probably be presented to other companies, minimizing the marketing value, which should have been generated due to the flyer. Therefore, the company not only didn't gain any value, but it lost value and maybe even a competitive advantage.
This is one of the main reasons why AiSEC was developed, so private data stays private, without having to miss out on benefits artificial intelligence provides.
¶ What does AiSEC offer?
AiSEC offers several benefits for a company.
-
Multitude of Models:
While most proprietary artificial intelligence solutions, try to invent the one model that fits all, AiSEC offers various models, which perform especially well for different tasks. For example a model trained solely on sql will perform way better towards questions about sql than a model that was also trained for sql but mainly for various other topics. Therefore, our customers can choose which models they want to have available and can then choose the model for each task based on the topic of the question.
-
Data Safety:
The AiSEC solution is implemented on-premise therefore guaranteeing that no company specific data ever leaves the internal network or is leaked to other outside applications or entities.
-
Extensive Image and Video generation:
Due to a stable diffusion being the backbone of our AiSEC solution, this also offers more options for configuring and tweaking the outcome of an image or video generation. One not only can specify what should be depicted, but also what should not and one can do so using normal human language, simplifying the whole process. One can also upload a framework containing a specific layout grid to the model, telling the model to, for example, generate a whitepaper based on that framework, adding images only to specific positions in the grid.
-
Expandable Knowledgebase:
Through the use of RAG, retrieval-augmented generation, one can expand the knowledgebase of a model, without having to re-train it. Instead the information is added like a lexikon, the model can browse through. The model retrieves the data it is given, builds a vector database out of this collection of information, which it can then use if necessary.
This way also company-specific data can be added to our artificial intelligence solution, making it possible for AiSEC to browse through the provided company data and generate company-specific answers, images or perform other tasks based on company specific information. Knowledge can be added via files, which can easily be uploaded within a prompt or into a collection or via directories, which can be linked and synchronized via for example samba shares and can also be added to a collection.
Files which have been uploaded within a prompt, also called inline files, are user and session-based, meaning they can only be used for this specific session by this user. One can also upload PDF files. AiSEC will scan and evaluate them and then turn everything written into text and save all image data with all references, so it can retrieve those when asked for them. If one wants the model to use a specific collection for answering a question or performing a task, one only has to provide a "#" within the prompt, which will show a list of collections to choose from. One can also specify that all models should by default use the collections rather than searching for the answers in the web.
-
No Vendor Lock-in:
For AiSEC, only open-source solutions are used, which prevents vendor lock-in. This means that customers are not tied to specific providers or proprietary technologies, which often leads to long-term cost pitfalls and limited flexibility. With open-source solutions as its foundation, AiSEC customers benefit from the dynamic development of a global community that is open to innovation and adaptation to new technological trends.
This enables seamless integration of new services and features without the limitations of proprietary ecosystems. Furthermore, our customers can adapt their IT infrastructure to their needs without having to worry about compatibility with a specific provider by forgoing proprietary solutions.
-
Extensive Prompt Engineering:
The stable diffusion running in the background adds additional options for prompt engineering. Our solution does not only offer positive prompting but also negative. This means one cannot only specify what the answer should include, but also what should be excluded. This makes it possible to tweak the outcome of a prompt and therefore get more relevant answers to ones query than otherwise.
-
API Integration:
AiSEC offers an API and therefore the possibility to integrate this solution into different third-party tools such as visual studio code or jupyter notebooks. This makes it possible to use all of AiSECs functions within those tools.
-
Search Engine Integration:
AiSEC offers the possibility to integrate different search engines and to also configure them. This allows for a wider knowledgebase of a model, so it can more accurately answer any queries.
-
Workspace Configuration:
One can configure the behaviour of a model via a workspace using human language to set different parameters. One can first choose the model this workspace should use, then the knowledgebase the model should use and then set the purpose of the workspace as well as the restrictions for that workspace. For example one can define a workspace, which should act as a tutor for children between 11 and 14 years, which should never just do the work for the kid, but rather guide it through the process and give hints and tipps. Another example would be a workspace used for training purposes, which poses questions to check ones knowledge, without ever giving the answer, but rather comparing the answer provided with the correct one and calculating a score, which is shown at the end.
-
Guaranteed Functionality:
At Sphinx, we fully stand behind the quality and performance of our AiSEC solution. That's why we offer our customers a comprehensive functionality guarantee.
If ever problems should arise with the functionality of AiSEC that couldn't be prevented by our proactive maintenance, we work closely with our customers to identify the cause and develop a solution that minimizes disruption to their operations.
-
Full Management:
AiSEC is managed by the Sphinx team as a Managed Service. This means that our experienced team continuously monitors all components of AiSEC to ensure that every aspect of the solution - from infrastructure to applications - operates smoothly.
Through proactive maintenance and immediate response to potential issues, we minimize downtime and guarantee maximum uptime for critical workloads. With AiSEC, our customers can focus entirely on their core business, while we handle the behind-the-scenes tasks to ensure everything runs as expected.
-
7 x 24 proactive Monitoring:
A proactive approach is key to preventing problems from arising in the first place. That's why we place great emphasis on comprehensive monitoring of all system aspects for our AiSEC solution.
Our powerful monitoring system provides real-time insights into the performance, security, and integrity of the platform, enabling early warnings for potential issues, and allowing our team to take corrective action often before our customers even become aware of a problem.
-
Third-Level Support:
A top-notch support team is the foundation of any successful solution. We offer a Third-Level Support for AiSEC that goes beyond common service standards. While First- and Second-Level Supports often only cover general inquiries and basic technical issues, our Third-Level Support is specifically designed for complex, in-depth technical challenges.
Our experts on this level work directly with developers and system architects to develop individual solutions for particularly demanding problems. With this top-tier support, we ensure that even the most complex challenges become no barrier to success for our customers.
¶ AiSEC Basic Tariff
AiSEC is offered as a service, we provide the hardware, which will be installed on-premise and we offer different machine learning models. Which models will be available, will be defined after a meeting with our customer, where we consult for which purposes AiSEC is needed and therefore which models would fit best for these purposes. After a successful implementation, our customers can add and remove users independently by themselves. In the basic tariff 20 users are included. Additional necessary users can be easily added for a fee. Furthermore the basis tariff also includes a configuration by us, such as adding necessary knowledge bases and an initial training about AiSEC, so our customers can use the full potential of our solution. Not included in the basis tariff is a RAG implementation, however this can be added as well if the need arises.