5 New Product Features from Our Startups at Fraunhofer HHI

What can startups with access to world-leading research build? About a week ago, our startups gathered for their internal Portfolio Day at Fraunhofer HHI and pitch their progress to the room.

Here's an insider look at what our venture lab has been building:

1. Architects can now generate snazzy building renderings in seconds, with Projected's new AI tool‍

Projected's platform helps architects test whether a design is feasible before committing real money to it. Their new feature uses a proprietary AI engine, trained on nearly a million real floor plans, to generate facade renderings in about 20 seconds, work that would normally take a designer days.

2. Property teams can now "talk to" their buildings using their own AI, thanks to Kukan AI

KUKAN AI turns physical spaces into digital models that property teams use to manage them. The new piece is a protocol called MCP (Model Context Protocol), a way for AI models like Claude or ChatGPT to plug directly into external data sources and act on them. With it, users skip KUKAN’s dashboard entirely and just ask their own AI questions about the space, and the AI reasons over the real building data behind the scenes.

3. AI teams can now catch almost all malicious prompts in real time, with Layerwise

Layerwise's technology looks inside an AI model's internal layers while it's actually running, rather than just checking its outputs after the fact, which is how most safety tools work. That approach lets it catch over 95% AUROC of prompt injection attempts against a 55% AUROC success rate for Meta's own PromptGuard 2, while using a fraction of the computing power competitors need.

4. Preclinical researchers can now test multiple species on different platforms, with iMouse

iMouse’s latest development expands the solution beyond mice towards multiple preclinical species. They do this by combining DigiFrame hardware and retrofit adapters for existing cage systems with the iMouse Hub, bringing continuous video, data and AI-based analysis into one platform. The goal with this shift is 24/7 unbiased remote observation while remaining compatible with existing laboratory infrastructure.

5. Vets can now see why an AI flagged a sample as cancerous, with testblu

testblu’s blood test uses AI to detect early signs of cancer in dogs. Their newest addition, built with Fraunhofer's Explainable AI (XAI) team, opens up the model's decision-making so vets can actually see which markers led to a result, rather than trusting a black-box output.

If any of these products caught your eye, reach out to the startups directly, most are looking to connect with early users, partners, and collaborators.

If you're an investor, our Demo Day on October 8th is your chance to meet several of them in person.