AI systems that follow from the evidence.
Illative is an AI consultancy. We help teams decide what to build with machine learning and language models, build it, and make it reliable enough to depend on.
What we do
Strategy & scoping
Where AI earns its keep in your business and where it doesn’t. A written, costed plan before anyone writes code.
Build
LLM applications, retrieval and agent systems, data pipelines, and integrations, shipped as software your team can run.
Evaluation & hardening
Test suites, failure analysis, red-teaming, and monitoring, so you know how the system behaves before your customers find out.
Selected work
Editing model behavior in the weights
A representation-engineering technique, introduced in 2024, that removes a model’s refusal direction directly from its weights, with no fine-tuning and no prompt overhead. Now a standard tool in the open-weights ecosystem, at over 2 million downloads a month on Hugging Face. It is also an interpretability method: the same direction lets you audit why a model refuses.
sordino.sh: code securely with AI
Masks secrets and sensitive information on your machine before a prompt reaches an AI coding tool. What it detects leaves only as opaque tokens, then is unmasked locally in the response. The inverse problem to abliteration, on the same boundary between a model and the world.
About
Illative LLC is a small, independent firm. The name is a grammatical term for the case that marks movement into something, and a logical one for words like “therefore”: a conclusion that is earned by what came before it.
Our work spans model internals through production systems. That is how we work: every recommendation traces back to your data, your constraints, and a measurement.
Contact
New engagements and general enquiries: contact@illative.ai