Sixteen pitches.
Three questions
each.

A sensor box plus software that watches rotating machines — motors, pumps, gearboxes, CNC spindles — and warns before they break. The difference they claim is causal: instead of saying “this vibration looks unusual”, the system says which physical fault is causing it (for example, shaft misalignment driving bearing resonance) and when it will fail. Target buyer is the German industrial Mittelstand. The company comes out of Constructor University Bremen research.

Their own site says the product is in development and not available to buy. The numbers on the page are market numbers, not customer results: 73% of bearing failures give no warning signal, ~€18,000 average cost of an unplanned failure at a German SME plant, ~9 hours of downtime per failure. Software MVP is described as ready, hardware is not. causenex.com, retrieved 21 Sep 2026 · earliest stage of the sixteen

Lead with this

A redesigned version of an existing, approved medicine. The drug class — an IL-1 receptor antagonist — already exists as Kineret/anakinra, which works but has a very short half-life (about 4 hours), needs a daily injection and a cold chain. Cytokyn M’s molecule, CIRPELIRA, is the same protein rebuilt in a different order so that it is physically more stable, which should mean fewer injections. The targets are Long COVID first and then “inflammaging” — chronic inflammation in people over 50.

From the deck: raising €2.5M pre-seed to finish preclinical work, file with the EMA and be ready for a first-in-human trial; Phase 1 planned to start 2027, Phase 2 in 2028. Semifinalist in the XPRIZE Healthspan competition. Co-founder Prof. Dr. Dr. Michael Wiechmann comes from Sandoz. The comparison table against Kineret is marked as in silico (computer), in vitro and bacterial-cell work — the advantage claims are modelling and bench data, not animals or patients yet. Cytokyn M pitch deck, slides 8–23 · deck read 21 Sep 2026

Lead with this

A “flight simulator” for salespeople: AI plays the customer, the rep practises the conversation, and the system scores things that show up in the P&L — attach rates, objection handling, upsell, margin. It also benchmarks reps and teams against each other, which is what a sales director actually buys. Main markets are retail chains, dealerships, pharma reps and telco.

The one real proof point on the site is a pilot with iSpace, an Apple Premium Partner inside the ASBIS group: +20% revenue per rep, +21% gross profit, 146% ROI, and ASBIS then committing to roll it out across the iSpace network in March 2026. Everything else — the “$237 basket gap”, “$120K enterprise deal gap” — is illustration, not customer data. No pricing is published. dealbooster.ai, retrieved 21 Sep 2026 · deck is image-only, no extractable text

Lead with this

Turns company knowledge into training material automatically. You upload a document, paste text or record your screen; the AI reads it, builds a structure, and produces a learning-optimised video with quizzes, either in Explaino’s own LMS or exported to the customer’s (SCORM). The pitch is onboarding, software rollouts and the knowledge that leaves when an employee leaves. Everything is hosted in the EU, with GDPR, ISO 27001 and DORA compliance pushed to the front — which is the real selling point for German banks and insurers.

Named results come from two big regulated German institutions: a VP at Sparkasse citing 7× higher knowledge retention, and an IT solution designer at AOK citing 85% faster content creation. The site’s ROI calculator will happily show “€1,039,980 annual savings” for a large headcount — that is a model, not a measured customer result. explaino.ai, retrieved 21 Sep 2026 · deck is image-only, no extractable text

Lead with this

An oral cancer drug that attacks the defence system tumours use to survive chemotherapy. Cancer cells protect themselves from oxidative damage; FerroptoCure’s compound FC-004 blocks two of those defences at once (xCT and ALDH), which pushes the cell into ferroptosis — an iron-dependent form of cell death. First indication is triple-negative breast cancer, where survival is poor and resistance is common. The science comes out of ~15 years of work at Keio University.

This is the most advanced company on the list. From the deck: $12.5M already raised in equity plus Japanese AMED grants, a collaboration with MD Anderson, three patents, 85% tumour reduction in mouse models, and — the key fact — a first-in-human Phase 1a/b trial already running in Japan (registry number jRCT2031230531, National Cancer Center), with “favourable safety profile observed” so far. Plan: Phase 1 in Australia 2027, Phase 2 in US/EU 2029, licensing to pharma along the way. FerroptoCure deck slides 6–11 · deck read 21 Sep 2026

Lead with this

An engine that does the parts of an AI system which must be exact. The pitch: today’s models guess, cost a fortune in GPUs and cannot explain themselves, which will collide with the EU AI Act. Holo Engine is presented as the opposite — same input always gives the same output, runs on a normal CPU in about 10 MB, learns from six to eight examples, and returns an audit trail. In practice it is sold as a tool the LLM calls: the model handles language, the engine handles routing, retrieval, verification and state.

Raising $3M at $30M post-money (10%), 18-month plan, 1–3 strategic investors. Benchmarks are self-run: 90% Recall@3 without embeddings, 33× faster than PySCF on a quantum-chemistry basis set, 13–39× on exact solving versus SymPy. The technical story rests on hyperbolic p-adic trees and energy minimisation, and is credited to a single inventor, Kirill Kazakov, with a physics/maths advisor. No customer is named anywhere, although the stage is given as product-market fit. Hi Holo deck + hiholo.ai · read 21 Sep 2026

Lead with this

A writing workspace built for students instead of a general chatbot. It finds real academic sources, helps build an outline, keeps the student writing rather than pasting, and adds the tools students actually worry about: citation formatting, plagiarism check, AI detection, a predicted grade and a tutor that quizzes them. The positioning is deliberately about academic safety — 70%+ of students say they are afraid to use ChatGPT for coursework.

The only company here with a clean commercial picture: $780K pre-seed raised in May 2025 from six VCs and angels, on track for $1M ARR by March 2026, gross margin improved to 83% by optimising API cost, free-user base up 10× in a year, and now raising $1.5M for distribution. Price is around $30/month, or $7.90/month billed annually. In their own survey 45% of subscribers prefer Litero to ChatGPT, 19% prefer ChatGPT, 37% cannot compare. Litero deck + litero.ai · read 21 Sep 2026

Lead with this

Memory that is written with light instead of electricity. AI chips spend most of their time waiting for data to arrive from memory; MemStera wants to remove that wait using all-optical magnetisation switching — ultrafast laser pulses that flip magnetic bits directly, with the data staying put when the power is off. The claim is sub-nanosecond access, “hundreds of times faster” than conventional memory, aimed at AI training and high-performance computing.

The physics is real and published academically — the company says it builds on decades of research. What is not stated anywhere public: the development stage, any measured device result, the fab or research partners by name, the funding raised, or the team beyond the founder. The website is three paragraphs long. memstera.com, retrieved 21 Sep 2026 · deck is behind a DocSend email gate

Lead with this

SEO for the age when customers ask ChatGPT instead of Google. nBlick builds buyer personas, fires thousands of simulated questions at the main AI assistants, and reports how often your brand is mentioned, in what tone, and against which competitors — then recommends what to change. Buyers are marketing and growth teams; there is a separate plan for agencies managing several clients.

Twelve-plus customer logos are shown (Anchorage Digital, Certideal, Macrocosmos, Brighthire, GuruWalk and others), which is real for a company this young. No revenue, user count or price is published — you must create an account to see pricing. The category is crowded and well funded: their own comparison pages name Profound, Peec AI, Otterly and Semrush. nblick.com (redirect from trynblick.com), retrieved 21 Sep 2026

Lead with this

A replacement for the diesel generator. Ammonia is easy to transport and already has a global supply chain; Neology’s box cracks it into hydrogen on site, cleans the gas and runs a fuel cell, giving power with no CO₂ at the point of use. Two products from the same core: a standalone cracker that sells hydrogen to whoever needs it, and the full power generator for off-grid and backup use in construction, telecom, agriculture and defence.

From the deck: seed of USD 2.5M closed in December 2025, a convertible loan open now, pre-Series A planned for Q2 2027; revenue of USD 300K in 2025 and USD 750K projected for 2026; pilot programme and commercial availability from summer 2026. Unit economics claimed: 55 cents/kWh versus 81 cents/kWh for a diesel generator, and CHF 5/kg of hydrogen cracked on site versus CHF 20/kg delivered by tube trailer. Investors include the Toyota Mobility Foundation, Venture Kick and Kickfund. Working prototype demonstrated on video; team has four PhDs across catalysis, reactors and process design. Neology deck + neology.ch · read 21 Sep 2026

Lead with this

During cancer surgery the surgeon waits, with the patient open, while a lab freezes and stains a piece of tissue to say whether the edges are clean. That wait is ~30–40 minutes, needs a pathologist who is often not there, and destroys the sample. PolaSight puts the fresh tissue in a standard cassette, scans it with polarised light for 60 seconds and returns an AI map of healthy versus tumour tissue. Nothing is stained, frozen or consumed, so the sample survives for normal pathology afterwards.

The most complete commercial plan in the cohort: hardware at ~$30K against competitors at $200–400K, plus $150 per scan and $8K/year service; claimed hospital saving of $1,580 per case. Built with the University of Bern and Inselspital; 2 patents filed through the university with 6 more in preparation; 6 peer-reviewed papers; 1,711 biopsies acquired in 18 months. Accuracy shown as 97% pancreas, 98% liver, 94% lung, 91% breast — but the study sizes beside them are N=19, N=11, N=8, N=9. Status today is research use only; CE mark targeted Q4 2026. Closing CHF 500K pre-seed, then a $3M seed. PolaSight deck slides 7–16 · read 21 Sep 2026

Lead with this

AI models that run on quantum computers instead of GPUs. The argument: transformers are hitting a wall — cost grows quadratically with context, energy use is becoming impossible, and benchmark gains are flattening. Query Machines builds quantum versions of attention and transformer architectures, aiming at reasoning tasks rather than chat, and eventually a pay-per-token API where the customer never knows the hardware underneath.

Sitting inside Europe’s strongest quantum cluster — TU Delft/QuTech, YES!Delft, a joint research project with Prof. Sebastian Feld’s quantum machine learning group. They say every claim has run on real quantum hardware: IBM, IonQ, IQM, Rigetti and Quantum Inspire, at about 25 qubits today, with the roadmap needing 64, then 100, then 1000+ for the actual product. Five whitepapers, three validated proofs of concept. Both founders are described as MSc students in applied physics; raising a pre-seed, use of funds 45% R&D team, 45% compute. Query Machines deck slides 5–11 · read 21 Sep 2026

Lead with this

A rocket that leaves its energy on the ground. A large microwave beam from a ground array heats hydrogen on board, so the vehicle carries propellant but not fuel energy — on their numbers, 55% propellant instead of 90%, payload ~20% instead of 4%, single stage, reusable a thousand times, and no explosive fuel on board. The long game is launch at $40/kg for the orbital data centres that NVIDIA, Google and Starcloud are already building. The near game is much smaller: electric satellite thrusters now, anti-drone directed energy in 2028–29.

From the deck: a propulsion lab running with Turin Polytechnic in Tashkent, a $200K satellite propulsion contract signed, gyrotron access secured through Bridge12 Technologies (an MIT spin-out), an engineering partnership with Roketsan, peer-reviewed papers, and four patents to be filed. Raising $1M at $10M post-money on a SAFE, with $550K committed; the milestone that unlocks the $10M+ seed is a 700-second specific impulse hydrogen thruster demo. Advisors include a gyrotron specialist and a NASA Mars landing GNC professor. Every component is claimed to exist at TRL 8–9 — the invention is the integration. Spaceborne deck · read 21 Sep 2026

Lead with this

Robots cannot learn from video, because video has no objects and no physics. Splatica takes a three-minute walk-through filmed on a consumer 360° camera and returns a simulation-ready copy of that room: photorealistic, with each object separated and named, collision shapes, mass and friction, exported in the format NVIDIA Isaac Sim expects. A robot can then be trained in the copy and deployed in the real room the next day.

From the deck: $120K early revenue, 2,000+ environments already scanned, official technology partner of Insta360, NVIDIA Inception member, and one proven zero-shot deployment with robotics startup Opteran. Cost claim: ~$15 per scene against $1,000–$5,000 for a professional LiDAR survey. The team has worked together for ten years across WayRay, Humanoid and Jaguar Land Rover. Business model is two-stage: digital twins pay the bills now, the scene library is the asset later. Splatica deck slides 4–11 · read 21 Sep 2026

Lead with this

Batteries lose capacity mainly because usable lithium is consumed over time. This team injects a proprietary electrolyte — they call it a lithium inventory refill agent — into the cell without opening it, and claim the pack comes back to roughly its original capacity. If it works at scale it turns a replacement into a service: the pack goes back to work instead of to recycling.

Claims on the site: ~99% of original rated capacity restored, at $15 per kWh, about 4× cheaper than replacement, non-destructive, nothing disassembled. Target buyers are fleet operators and grid-storage owners. The founding team previously built HyPoint and HyWatts, with advisors from Tesla and Electric Hydrogen. What is not published: how many cells have been treated, how they behave after treatment over time, or any third-party verification. thelithium.com, retrieved 21 Sep 2026 · stage given as proof of concept

Lead with this

A single workplace for university teachers, replacing the pile of tools a professor juggles today — Moodle, Zoom, Mentimeter, Excel, Notion and ChatGPT. It runs the live class, builds the course, generates exercises and rubrics for the professor to approve, shows who is falling behind, and gives students a tutor bounded by the syllabus. Co-founded by a strategy professor at HEC Lausanne, which is why it is designed to be bought by one professor rather than by a university IT department.

Pilots at HEC Lausanne, University of Geneva, Durham, the IOC and LUMS Pakistan, with 1,000+ learners; incubated at Qatar Science and Technology Park. Claims: 122 hours per semester recovered, 8× more in-class exercises, 4–6 hours saved per assignment cycle, set up in under 30 minutes with no IT approval. GDPR and FERPA compliant, no training on faculty data, Moodle-compatible, audit trail on every AI action — the governance answer universities ask for first. tortin.com, retrieved 21 Sep 2026 · stage given as MVP

Lead with this