GravityLab launches AI assistant for DonateLife Victoria

GravityLab has launched an AI assistant for DonateLife Victoria that gives staff cited answers drawn from the organisation's own documents.

Called Sal, the assistant is designed to help clinical and operational staff find specific information during time-critical organ and tissue donation cases. Staff often need a single fact quickly, such as a hospital location or contact detail, and previously had to search manually through lengthy documents or contact specialists after hours.

Tony Holland, General Manager at DonateLife Victoria, said the system had already changed how quickly staff could retrieve information during active cases.

"That's the kind of question our staff deal with constantly, and it used to mean stopping everything to search," Holland said.

"Now it's instant. That difference matters when someone's in the middle of a case and every minute counts."

The assistant was built on Salesforce Agentforce and searches 26 clinical and operational documents each time a user submits a question. It returns an answer with a source reference so staff can check the original material rather than rely on the tool alone.

How it works

The system is limited to approved internal documents rather than general web-based knowledge. It retrieves a relevant document, checks whether the material addresses the question, and cites the exact source for the answer.

This approach is intended to reduce the risk of fabricated responses, a persistent concern for organisations using generative AI in settings where staff make operational or clinical decisions. The assistant is also designed to decline to answer when the information is not available in the source material.

Holland said confidence in the source of an answer was central to the project.

"We needed the nurses to know with a high level of confidence that the answer that's been given is accurate, and where it's come from. That's where the guardrails are being built," Holland said.

Table problem

A key challenge involved data held in tables inside SharePoint documents. Many important facts, including hospital details, addresses and contact information, were stored in tabular form, which can be difficult for AI search systems to interpret consistently.

To address that, GravityLab built a process that reads tables in key documents and rewrites each row into markdown so the assistant can retrieve the information more reliably. The source documents remain in SharePoint, allowing DonateLife Victoria's clinical teams to keep publishing and updating files through their existing workflow.

Rather than relying on informal testing, GravityLab assessed the assistant using real questions with verified answers. The process highlighted weaknesses in standard document search and showed that the table-handling method improved retrieval.

Time savings

DonateLife Victoria said the gains could add up across a single case, where multiple information requests may arise in quick succession. According to Holland, staff had previously spent around 15 minutes on some interactions.

"It might be 15 minutes per interaction, but there can be numerous interactions per case. That could save an hour or two per case," Holland said.

The system went live in production in September 2026 and is now being rolled out across the wider clinical team. Staff continue to maintain the source material in SharePoint.

Daniel Howell, Founder at GravityLab, said the project reflected a broader emphasis on limiting AI responses to information that can be verified.

"The easy part of AI is making something sound confident. The hard part is making it honest, willing to say 'I don't know' instead of guessing. That's what we built into Sal, and it's the same standard we hold on every Salesforce project we deliver. Proven against real data before it goes anywhere near someone making a real decision," Howell said.