dennis.law – doing what Dennis Denuto wasn’t able to

“I have an aviation background and tell people that The Castle is like Top Gun for lawyers,” Willson said. “The irony being dennis.law can argue constitutional law eloquently and do everything Dennis Denuto wasn’t able to do. Except it probably also cannot fix the photocopier.”

dennis.law is used by dozens of law firms and barristers in Adelaide and is now expanding nationally.

In conversation with Lawyers Weekly, Willson said the platform “excels” at Australian legal research, and that it had not seen anything else on the market able to do a thorough job of it.

He said that the best output it had seen from competitors or basic large language models was a handful of basic principles and one or two High Court references. By comparison, dennis reads through hundreds of cases in detail and finds supporting citations for every legal proposition, often producing a list of 30 to 50 references.

The platform also frequently surfaces on-point legal principles from judges’ reasons in unrelated cases, the provider said.

The second problem he said it addresses is the time practitioners spend writing prompts. In dennis, prompts are optional and offer high-level guidance only, according to the provider. Users upload matter documents, choose the output type – such as an opinion or written submissions – say which side they represent and click go.

VIEW ALL

“This frees up a lot of time for professionals to do other tasks rather than spend all day wrangling their LLM of choice,” Willson said.

Why general-purpose tools fall short

Asked why “vanilla ChatGPT” fails to meet these needs, Willson said the answer was technical and involved several inherent limitations.

The first is working memory. He said the context window of such models is about 1 million tokens, representing around 500,000 words, but that in practice, only about 250,000 words are usable. A 500-page brief is already multiples of that, he said, so a single model cannot read everything before research even begins.

Willson also said general models lack access to source judgments and legislation, and instead run web searches and read blog posts from law firms.

“This is the wrong approach and instead of reasoning from first principles of judgement reasons, they are repeating some lawyer’s opinion,” he said.

Further, Willson said, such models cannot produce first-class .docx outputs with native formatting, and do not check their own work. In an attempt to satisfy a user’s prompt, he said, they will often “fill in” detail, a problem commonly referred to as hallucination.

Dennis, he said, works in the opposite direction. The platform has access only to cases and legislation, and is deliberately blocked from the internet and run in its own isolated sandbox for both security and quality.

Using search tooling the provider developed, dennis fills its working memory only with cases and specific paragraphs of judgment reasons that are highly relevant, according to Willson.

He said the system is multi-pass and uses 20 stages in its production pipeline, drawing on multiple AI agents with different purposes, as well as traditional computer science steps and other local machine learning models.

“It often surprises people that most of dennis.law is actually traditional but advanced code and tooling,” he said. “Dennis is the opposite of a thin wrapper around ChatGPT; it is a completely customised system comprising of hundreds of detailed parts all tuned for Australian legal jurisdictions.”

Local models and early uptake

Reflecting on the trends the platform is watching, Willson pointed to open-source models that can be run locally. A lot of the platform already runs on on-premise hardware, he said, and it would ultimately like to see large models equivalent to Opus 5.5 and GPT-6 run on its own hardware.

He said it was encouraged by the uptake from its soft launch into the Adelaide market, reporting almost 100 per cent conversion from trial to paid subscriptions. Usage and feedback have been very strong, he said, with firms “running dennis hot” during both trial and paid plans.

dennis.law uses a product output pricing model, which Willson said makes the platform essentially free to sign up to, with costs incurred only when users need thorough, sophisticated legal research that other platforms fail to deliver.

He added that it was beginning to see new sign-ups through word of mouth, and that the platform hopes to see similar results at Lawyers Weekly’s LawTech: AI Summit, being held on 30 October, for which it is a platinum partner.

“The feedback has been genuinely amazing, and we love hearing how helpful dennis has been for firms,” he said.

Wholly Australian operated, dennis.law drafts court-ready opinions, advice, and submissions for Australian lawyers, reasoning directly from legislation and case law to provide rigorous factual analysis and legal research.

Debra Spizzo of Debra Spizzo & Associates said: “We are a two-person female firm and most of our clients are on legal aid or victims of domestic violence. Dennis Law is helping us to provide essential legal services efficiently to our clients. We can now more fully advise our clients as to risks, costs and final outcomes prior to commencing any legal action. We are more confident in the outcomes our clients will achieve.”

“We sleep better at night (Thank you Dennis).”