Raleigh founder Patrick Steele is building MicroBioCol around a problem he has encountered firsthand as a scientist in the biotech industry: Artificial intelligence can make work faster, but strict controls around sensitive information and documentation make the technology difficult to integrate.
“The product we’re working on is CadeLume—an AI proxy gateway,” Steele said. “It shows the complete log of the record of work that [the artificial intelligence] is doing, and we think that’s beneficial for people who are working in biotech.”
Steele has worked in the industry for more than 15 years, including as a scientist at Labcorp. In his own work, he said even relatively simple tasks such as editing controlled documents can’t necessarily be handed off to AI because companies need safeguards around confidential information and a reliable record of what was done.
Cadelume is designed to provide that concrete, auditable trail.
“With MicroBioCol, you have an actual log that’s auditable to show you what your models are doing, so you can see the tool costs, you can see the usage, you can see the eco impact stats, you can see all those things together,” Steele said.
The MicroBioCol tech specs
CadeLume creates that record by sitting between a company’s AI system and the services it communicates with. Rather than replacing models or AI agents a company already uses, an engineer redirects them through CadeLume’s gateway, which records the activity before sending it to its intended destination.
“All our product does, CadeLume Gateway, it just sits in the middle between your endpoint,” Steele said. “The gateway will record everything through it, and then it’ll send it to the [intended] endpoint.”
That includes interactions between AI agents, calls to large language models and the external tools those systems use. Engineers can also place policies and limits around those activities, including usage and energy limits.
To make the record auditable, Steele said CadeLume links each recorded event using cryptographic hashes. Each new action becomes another link in the chain; if someone later alters a record, the chain no longer matches, making the change detectable.
“Every tool call, everything is connected to themselves,” Steele said. “If you try to modify that chain, then it breaks, and so an auditor can easily come in and say, ‘Well, this was modified; it’s not accurate.’”
Eco-Impact philosophy
Steele sees CadeLume’s environmental monitoring as a key differentiator from other AI gateways, reflecting MicroBioCol’s broader philosophy that AI should be deployed with its human and environmental costs in mind.
“We want AI to assist people, not replace people,” Steele said. “We’re trying to build tools that think about the environment and people.”
CadeLume is designed to track metrics including token usage and estimated energy and carbon emissions, while allowing companies to set limits around resource consumption. Steele sees that capability as particularly relevant for biopharma companies pursuing emissions-reduction goals.
For example, CadeLume’s tracking could show that an AI system is using an unnecessary number of calls to complete a task. A company could then streamline the process, reducing token usage and potentially cutting both costs and energy consumption.
The goal is to give companies visibility not only into what their AI systems are doing, but also the resources they consume—and the ability to constrain both.
Second Layer: CadeLume Work
The gateway is only the first layer of Steele’s vision. After validating it, MicroBioCol plans to build Cadelume Work, an employee-facing layer designed to let workers use AI agents within the controls and audit trail established by the gateway.
Steele gave the example of an AI agent editing a regulated method. The agent could complete the task, return the document to an employee for review and sign-off, then move it to the next reviewer—all while CadeLume records the process.
“Once it’s all completed, then I can submit the method for release, and it can be approved,” Steele said. “So all those steps—that’s what CadeLume Work is going to do eventually.”
QUICK BITS
Startup: MicroBioCol AI
Founder: Patrick Steele
Founded: 2025
Team size: 4
Location: Raleigh
Website: microbiocol.com
Funding: Raising pre-seed
MicroBioCol has reached the MVP stage with CadeLume and was accepted into the NC SEA Change cohort. The company is now looking for design partners, particularly small- to mid-sized biopharma companies willing to test the gateway on a specific workflow and help shape the product.
Within the next year, Steele hopes to secure at least three design partners and three paying customers. Biopharma is the starting point because of his industry experience, but he eventually hopes to expand CadeLume into other regulated sectors adopting AI, including finance, local governments and government agencies.
