AE Studio

AI and ML

AI Agents

Agentic automations — orchestrated AI agents that handle complex workflows and back-office processes in production environments.

AI and ML

AI-Native Products

End-to-end builds of AI-native products: conversational and voice experiences, personalization, copilots, generative features, and entirely new product categories — from first idea to shipped production.

AI and ML

Custom ML Models

Purpose-built machine learning models for client environments, including predictive models, neural decoders, ranking, vision, and forecasting.

Data and analytics

AI Knowledge Management

Knowledge graph and data foundation work that structures the data underneath AI systems, enabling reliable agentic automations and intelligence layers.

Advisory and strategy

Alignment Research

Frontier AI alignment R&D pursuing neglected approaches — pretraining alignment, neuroscience-inspired alignment, empirical AI consciousness research, and alignment policy — with collaborators from Anthropic, DARPA, Princeton, Redwood Research, and Los Alamos National Laboratory.

A pretraining alignment method developed with Anthropic that isolates dangerous knowledge into switchable modules, enabling deployment-time access control (e.g., virology knowledge on/off per authorized user) while maintaining model performance.

Our assessment: Peer-reviewed research method co-developed with Anthropic; clearly productized concept with a named methodology and published paper.

A mechanistic interpretability method for self-interpreting LLM embeddings, published as a peer-reviewed paper.

Our assessment: Named, peer-reviewed interpretability technique with a direct paper page; substantive research IP.

A mechanistic interpretability research finding on steering resistance in LLMs, published as a paper.

Our assessment: Named research output with a direct page; substantive but described only briefly in the captured text.

A suite of AI-native learning products shipped into real classrooms for Alpha School, constituting a hyper-personalized AI learning platform.

Our assessment: Client-built IP cluster with named products and real-world outcomes; substantive but belongs to the client, not AE as reusable IP.

Production software and neural decoders for Blackrock Neurotech's brain-computer interface that restores movement and communication in paralyzed patients.

Our assessment: Named delivered platform with clear scope; client-facing product rather than reusable AE framework, but prominently featured.