Research
Where the thinking lives.
We maintain long-running research threads that feed directly into the systems we ship. Below are the questions we are working on this year.
Enterprise Legal AI
Citation-grounded reasoning over long, structured, multilingual documents.
We are studying how to combine retrieval-augmented generation with formal-document structure so that a legal model can give answers a counsel can stand behind.
AI Governance
From policy papers to operational telemetry — making governance real.
Governance is often written as principles. Our work turns those principles into auditable logs, evaluation harnesses, and architectural constraints that ship with the system.
Retrieval-Augmented Generation
Indexing strategies that survive the messy reality of enterprise content.
We treat RAG as an indexing and retrieval problem first, and a generation problem second. The interesting failures live in the index.
Agentic AI
Bounded autonomy with continuous human oversight.
An agent should not surprise the operator. We design agents whose every action is observable, interruptible, and reviewable after the fact.
Cybersecurity AI
Detection that augments analysts rather than replaces them.
We treat the SOC analyst as the customer. Our research is measured by how much time it gives back, not by how many alerts it suppresses.
Large Language Models
Adapting frontier models to enterprise constraints.
Fine-tuning, distillation, and prompting are tools, not destinations. We choose based on the deployment envelope: latency, privacy, cost, and the cost of being wrong.
Oil & Gas AI
Industrial telemetry, asset reliability, and operations under uncertainty.
Decades of domain knowledge in oil and gas still live in spreadsheets and shift handovers. We work on systems that respect that history instead of pretending it does not exist.