Technology
Our strategy is to use existing libraries, services, and APIs wherever possible, selecting the technologies best suited to each application. We develop proprietary software and infrastructure only where necessary.
Open-Weight LLMs
We work with open-weight model families from providers and research organizations such as Moonshot AI/Kimi, Mistral AI, Meta/Llama, Alibaba/Qwen, DeepSeek, Google/Gemma and Ai2/OLMo. They support flexible customization and deployment in cloud, private and sovereign environments. Licensing, available source material and reproducibility vary between model families.
AI Application Engineering and LLMOps
Technologies for retrieval-augmented generation, vector search, agent orchestration, evaluation, observability and scalable deployment, including Supabase, PostgreSQL/pgvector, MLflow, Docker and Kubernetes.
OpenAI
Frontier models for reasoning, coding, multimodal applications and agentic workflows. APIs support text, real-time voice, images, audio, embeddings, content moderation and tool-enabled AI agents.
Anthropic
Claude models and APIs for reasoning, coding, document analysis, multimodal applications and tool-enabled agents, available directly and through major cloud platforms
Google Cloud AI
Gemini models and a broad Model Garden, combined with managed services for machine learning, AutoML and AI agents. Tight integration with Google Cloud data services, scalable GPU and TPU infrastructure, and production-grade model deployment.
Amazon Web Services
Amazon Bedrock provides managed access to leading foundation models and services for agents, knowledge bases, model customization and guardrails. Amazon SageMaker AI supports the complete machine-learning lifecycle, backed by scalable GPU infrastructure and AWS-designed AI accelerators.
Microsoft Azure
Azure AI Foundry and Azure OpenAI provide managed access to foundation and reasoning models. Copilot Studio and Azure AI services support enterprise agents, conversational applications, workflow integration and human-in-the-loop processes, with scalable accelerated computing infrastructure.
Python Machine-Learning Ecosystem
scikit-learn for classical machine learning, preprocessing and evaluation; PyTorch,
TensorFlow and Keras for scalable deep learning and production deployment;
Hugging Face for pretrained and multimodal models; and
Jupyter and Google Colab for interactive development, prototyping and collaboration