AI service as workflow acceleration. As a technical AI service provider, we connect and enrich your systems and employees with relevant new explainable knowledge. Cross-departmental, suitable for agile organizational forms.
HighPots offers you the possibility to make all information and data from all your systems accessible via AI. Furthermore, we also show you what information you should not make accessible to any AI system. Or only to certain user groups.
At HighPots, we implement the use cases described above using a method called RAG. Retrieval-Augmented Generation (RAG) is an advanced technique in the field of artificial intelligence that enhances the utility of Large Language Models (LLMs).
RAG combines the text generation capabilities of LLMs with the ability to specifically retrieve information from external data sources and incorporate it into answer generation. This allows LLMs not only to draw on their trained knowledge but also to utilize current and specific data.
We represent your company’s knowledge in vector spaces/vector databases. You determine the storage location of these vector databases. The majority of HighPots customers store these vector databases in private clouds or on their own servers. This is often valuable knowledge (Intellectual Property) that gives the company a competitive advantage.
HighPots offers further security-relevant AI hybrid solutions. For example, the operation of powerful LLM AI models on your own servers. Only these have access to security-relevant data in the vector databases. We connect less critical systems for you with, for example, ChatGPT via API. Should tech corporations change their terms and conditions and use your company’s data, your Intellectual Property would be safe.
More than 400,000 internal wiki pages and more than 10,000 whitepaper PDF pages as a data basis. With a factual accuracy of 83% and high coherence with few logical breaks in longer texts, HighPots’ GenAI hybrid model convinced us. The company’s internal knowledge on vector databases combined with an on-prem Llama model and the API to the ChatGPT cloud. Data protection compliance and the protection of our Intellectual Property, combined with a low error rate, were our most important requirements.
Team leader Data Science
Astara Madrid
With the development of Analytical Intelligence, we at HighPots conscientiously integrate large amounts of data. Analytical Artificial Intelligence thus precisely reveals what is hidden in your data and provides well-founded forecasts and recommendations for action. Although Analytical AI is also based on an artificial neural network, it can be precisely understood and explained how the logical conclusions are reached. HighPots, as a Data Science service provider, develops Analytical Intelligences to identify correlations and optimize processes. As an AI service provider, after developing Analytical AI systems, we often connect them with generative AI so that HighPots customers can query the knowledge from Analytical AI in natural language. High-quality data is advantageous. Data qualities are results of the Digital Transformation.
HighPots develops expert systems for diagnoses, decision-making, and knowledge representation. Thanks to Bayesian statistics, even when little data is available. Especially in the fields of medicine, engineering, or finance. AI expert systems developed by HighPots also serve, for example, to analyze markets. Companies that want to investigate entering a market for which little data is available use HighPots’ AI expert systems.
Our Artificial Intelligence service includes Machine Learning networked with generative AI services. On your server systems as AI On-Premises, in our AI Open Source Cloud, in your cloud instance with third-party providers like Microsoft Azure, Google Cloud, or Amazon AWS Cloud. For this purpose, we additionally offer internal AI SaaS services in the HighPots Open Source Cloud.
HighPots offers fast integrations with ready-made generative AI models. “AI-ready” refers to pre-trained and fine-tuned AI systems; for example, OpenAI ChatGPT, Google Gemini, or Google Cloud Services Platform (CSP) for artificial intelligence on-prem on your own servers. These systems only need to be trained with your company’s knowledge. HighPots supports you with AI implementation and AI training. We can do this because we have experience in AI-relevant software development. This type of AI integration is often used for chatbots or for internal company claims management.
2D
Digital
Im Auftrag eines international-agierenden Medizingeräte-Herstellers entwickelten wir ein Expertensystem für MRT-, CT- und Röntgen-Geräte. Die Bilder aus diesen bildgebenden medizinischen Geräten werden von 2D in 3D umgewandelt.
Entscheidung der Deep Learning-Architektur: Verwendung eines CNN (Convolutional Neuronal Network)
Auswahl CNN: Aufgrund der Parameter Netzwerkbreite, Netzwerktiefe und Netzwerkauflösung nutzen wir das CNN EfficientNet
Pretraining: Datensatz-Architektur angelehnt an ImageNetvia, erstellen eines Datensatz basierend auf von Radiologen validierten Bildern
Finetuning mit TensorFlow: Testen des Datensatzes mit "ImageNet Large Scale Visual Recognition Challenge", danach Finetuning des vortrainierten EfficientNet mit TensorFlow (dazu wurden zuerst diejenigen Schichten definiert, die eingefroren bleiben sollen.
Reward Modelling und Reinforcement Learning: Erstellung von Agenten für Bilder-Klassifizierung
Ergebnis-Bewertung nach erster Iteration des Reward Modelling und Reinforcement Learning: Ergebnisse des Agenten von Radiologen bewertet (positiv/negativ)
Modellanpassung und teilweise erneutes Training
Anpassung des Agenten beim Reward Modelling und Reinforcement Learning: Hinterlegung eines komplexeres Bewertungsschema (Punkte-System)
KI-Modell-Verfeinerung - Reinforcement Learning: Erneute Ergebnis-Konfrontation mit den Radiologen
Wiederholung Punkte 5 bis 9: Erhöhung der Wahrheits-Eintritts-Wahrscheinlichkeit der Ergebnisse; Definition wahr (korrekt), wenn Fehlerquote SixSigma erreicht wurde
Entwicklung einer Schnittstelle zur Diagnose-Software des Geräte-Herstellers
Verbindung des KI-Experten-Systems mit generativer KI; Abfragen und Hinterfragen der Ergebnisse in natürlicher Sprache durch Ärzte und Patienten.
Generative AI was unable to achieve our 4-sigma error rate in AI early diagnostics. HighPots was the only service provider to achieve an error rate of less than 6.2 errors per million diagnostic cases in our imaging procedures with predictive and classification models.
Product Development
Philips Healthcare Eindhoven
As a service provider for Artificial Intelligence, we expand the knowledge of B2B AI systems in real-time. Through our methods in Machine Learning, we continuously increase the quality and entry probabilities of AI systems. For this, we use stochastic methods and sophisticated algorithms for supervised learning. We draw on deep knowledge in statistics and data science as well as many years of experience in software development.
Transformer models that do not offer natural language processing (NLP) can provide enormous added value for our customers. At HighPots, we develop solutions that process sequences in transformer models in such a way that, for example, reliable financial analyses can be performed.
Start generating real, demonstrable business transformations with the help of AI technologies, together with our AI experts.
Create a strategy with HighPots AI consultants to build your optimal data foundation. This spans the entire Data Science and Data Analytics lifecycle. Scale your AI workload with us.
Together with our AI specialists in AI governance, you can legally control, monitor, and maintain AI systems. Integrate AI responsibly into your entire business operations and combine generative AI, analytical AI, and machine learning. By identifying requirements and process data through business analyses and smart Requirements Engineering, errors can be avoided early on. HighPots can help you with a variety of AI use cases.
