Meet 7 Deep Tech Startups from Fraunhofer HHI’s Venture Lab

BERLIN, October 2026

Demo Day, Silicon Allee venture lab at Fraunhofer HHI

Early-stage deep tech usually carries two huge investor headaches: science that only works on paper, and cap tables wrapped in red tape. Our Venture Lab at Fraunhofer HHI fixes both.

For the past 1–2 years, our founders have worked alongside Fraunhofer HHI’s world-leading researchers, putting in 15,000+ lab hours, building IP moats, and securing paying enterprise customers, all while keeping cap tables pristine through virtual shares.

Last week, eight of our sharpest teams took the stage at our annual private Demo Day. From pitching live out of a cab in London to detecting dog cancer before symptoms appear, here are the eight startups in pitch order, the bottlenecks they solve, and their open rounds:

1. Anita - Hallucination-Free Legal Retrieval Grounded in Case Law

• The Problem: Legal AI tools hallucinate up to 33% of the time, which is a major liability risk, while 99% of German court rulings remain undigitized.
• The Solution: Strictly grounded Retrieval-Augmented Generation (RAG) powered by Fraunhofer HHI’s explainable AI. Every generated sentence links directly to verified court citations, cutting research time by 75%.
• Traction & Status: €16k MRR across 750 law firms, with high-margin enterprise APIs serving leading legal publishers.

2. iMouse - Automated Digital Biomarkers for Preclinical Drug Screening

• The Problem: 90% of drug candidates fail during clinical phases, largely due to intermittent, manual animal monitoring that introduces severe observational bias.
• The Solution: A retrofit vision-AI camera and sensor hub that continuously monitors lab animals 24/7 in their home cages, extracting objective digital biomarkers without handling stress.
• Traction & Status: Deployed across 9 institutions, shipped to the UK and Nordics, and contracted with a global pharma leader.

3. Projected - Step-Zero Generative Feasibility for Architecture

• The Problem: Evaluating residential building feasibility takes weeks of manual BIM drafting and fragmented environmental modeling to reach a basic go/no-go decision.
• The Solution: A multimodal 3D AI engine running real-time cloud simulations for wind, daylight, and energy from just an address, delivering feasibility studies 8x faster at 10x lower cost.
• Traction & Status: Live across 6 European countries with 120+ beta users and 150+ reports generated.

4. TwinCloud - Spatial Intelligence for Existing Buildings

• The Problem: Decarbonizing Europe’s ~110M buildings by 2045 requires massive renovation planning, but owners sit on fragmented, incomplete building data, and LiDAR capture costs €25k+ per unit.
• The Solution: Proprietary Gaussian Splatting technology, developed with Fraunhofer HHI, turns easily captured 360° camera footage into measurable 3D digital twins. AI then extracts measurements, insulation areas and heat-pump retrofit data for operations and renovation.
• Traction & Status: €0 → €100k revenue in 6 months; piloting with 6 major German real estate companies with 7 more pilots and demos lined up. Processing volume up 7x in one month (46 → 335 twins).

5. testblu - Early Multi-Cancer Screening in Canines via Explainable AI

• The Problem: 1 in 4 dogs dies of cancer, often because symptoms appear too late for effective clinical intervention.
• The Solution: CaniSense pairs liquid biopsy blood tests with DNA sequencing and Fraunhofer HHI’s explainable AI to detect minute circulatory tumor signals and give vets transparent diagnostic reasoning.
• Traction & Status: Outperformed benchmarks on academic data and secured 10+ signed clinic LOIs.

6. QuSine - Ultra-Low Noise Signal Generators for Next-Gen Radar & Sensing

• The Problem: Standard RF oscillators (quartz and atomic clocks) produce phase noise sidebands that drown out weak radar returns from small, slow-moving airborne targets.
• The Solution: Modular 19-inch RF signal generators delivering up to 100x lower phase noise than market alternatives, with a roadmap to miniaturize the tech onto photonic/RF chips.
• Traction & Status: Demonstrated live at European Microwave Week; 2 granted patents across 8 jurisdictions.

7. layerwise - Real-Time Internal Token Inspection for Frontier AI

• The Problem: Guardrails and external LLM judges add heavy latency without preventing internal exploits, while lab interpretability incurs a 200x compute overhead.
• The Solution: Powered by Fraunhofer HHI’s Layer-wise Relevance Propagation, Layerwise inspects internal token signals at inference time, delivering 100% prompt injection defense at only 25% compute overhead (heading toward <5%).
• Traction & Status: Tested on models from 3B to 500B parameters; selected for Creative Destruction Lab and presenting at NeurIPS.

‍

💬 Investors or partners: want an intro? Write to us at: team@siliconallee.com