Blab Tech
Transformer Lab Unveils Primus, an Autonomous Research Tool Claiming PhD-Level Output
On October 8, 2026, a quiet startup in the Kitchener‑Waterloo tech corridor announced a tool that could rewrite the pace of academic publishing. Transformer Lab, a 2024‑founded AI research company, revealed Primus – a system that, according to the founders, can produce research papers with a depth ranging from Masters‑level to PhD‑level within hours or days.
The company was founded by Ali Asaria and Tony Salomone, both of whom previously worked at Tulip and Well.ca. Earlier this year, Transformer Lab secured seed capital from Toronto‑based Ripple Ventures and Kitchener‑Waterloo‑based Garage Capital. With a core team of four, the founders have spent the past two years dissecting the workflow of researchers and engineering software to automate those steps.
Primus is not a single model but a constellation of AI agents, each calibrated to a distinct task. Some agents run on lightweight, cost‑effective models; others leverage frontier architectures from major providers or custom models trained in-house. Together, they conduct literature searches, formulate hypotheses, design and execute digital experiments, and compose a structured manuscript.
"A single prompt‑based research question can be transformed into a completed paper in hours to days without further human direction," the company says. Over a 30‑day period, Primus reportedly produced 30 papers spanning disciplines such as materials science, physics, seismology, protein design, audio generation, and 3D vision.
One of those papers has already been cited by a Google DeepMind publication, a detail noted in the reference list of the paper. Transformer Lab has not disclosed how the papers were evaluated for the claimed level of rigor.
Primus is free to start, but the startup is admitting users on a waitlist that is capped by the compute capacity available to the team. "We will add users as resources permit," the company said.
In a statement accompanying the launch, Transformer Lab clarified that it would prohibit users from submitting Primus‑generated papers to academic journals or from requesting volunteer reviewers. The policy is intended to prevent the tool from being used to bypass the peer‑review process.
Skepticism has already surfaced. Alessandra Buccella wrote for The Conversation that the very existence of science as an authoritative source of knowledge depends on human involvement.
Asaria has framed Primus as a potential remedy for the global shortage of machine‑learning engineers. "It could help researchers who cannot hire top talent to access a large pool of AI models at the touch of a button," he said. He also casts the tool as a step toward a new era of research that could accelerate progress across many scientific fields.
At present, Transformer Lab remains a small startup with a waitlist‑based rollout. No regulatory filings or court actions have been reported. The company’s next steps will likely involve scaling compute resources, refining the tool’s output quality, and monitoring academic community responses to AI‑generated research.
The launch of Primus marks a notable development in the growing field of AI‑assisted scientific inquiry, but its long‑term impact on research quality, peer review, and academic publishing remains to be seen.
The company was founded by Ali Asaria and Tony Salomone, both of whom previously worked at Tulip and Well.ca. Earlier this year, Transformer Lab secured seed capital from Toronto‑based Ripple Ventures and Kitchener‑Waterloo‑based Garage Capital. With a core team of four, the founders have spent the past two years dissecting the workflow of researchers and engineering software to automate those steps.
Primus is not a single model but a constellation of AI agents, each calibrated to a distinct task. Some agents run on lightweight, cost‑effective models; others leverage frontier architectures from major providers or custom models trained in-house. Together, they conduct literature searches, formulate hypotheses, design and execute digital experiments, and compose a structured manuscript.
"A single prompt‑based research question can be transformed into a completed paper in hours to days without further human direction," the company says. Over a 30‑day period, Primus reportedly produced 30 papers spanning disciplines such as materials science, physics, seismology, protein design, audio generation, and 3D vision.
One of those papers has already been cited by a Google DeepMind publication, a detail noted in the reference list of the paper. Transformer Lab has not disclosed how the papers were evaluated for the claimed level of rigor.
Primus is free to start, but the startup is admitting users on a waitlist that is capped by the compute capacity available to the team. "We will add users as resources permit," the company said.
In a statement accompanying the launch, Transformer Lab clarified that it would prohibit users from submitting Primus‑generated papers to academic journals or from requesting volunteer reviewers. The policy is intended to prevent the tool from being used to bypass the peer‑review process.
Skepticism has already surfaced. Alessandra Buccella wrote for The Conversation that the very existence of science as an authoritative source of knowledge depends on human involvement.
Asaria has framed Primus as a potential remedy for the global shortage of machine‑learning engineers. "It could help researchers who cannot hire top talent to access a large pool of AI models at the touch of a button," he said. He also casts the tool as a step toward a new era of research that could accelerate progress across many scientific fields.
At present, Transformer Lab remains a small startup with a waitlist‑based rollout. No regulatory filings or court actions have been reported. The company’s next steps will likely involve scaling compute resources, refining the tool’s output quality, and monitoring academic community responses to AI‑generated research.
The launch of Primus marks a notable development in the growing field of AI‑assisted scientific inquiry, but its long‑term impact on research quality, peer review, and academic publishing remains to be seen.
