An investment
worth playing with.
Development boutique showroom. Seven product brands connected by passerelles, three active patent families — a multi-product portfolio instead of single-device risk. Place a non-binding bid on any asset; Joeri Gredig reviews each one personally and replies with an offer.
Products
Readiness is read on a 1–9 scale per axis (as in the NASA TRL logic): RRL research · TRL technology · SRL system · IRL integration · BRL business · HRL human · LRL leadership. Seeking names who a brand is open to today — only two of the seven currently seek cash.
Strategic assets
Sponsoring partners — in-kind
Tier 0 — small & medium enterprises offering services, materials, or engagement without cash or equity. Lowest threshold, highest network signal.
Foundation partners — research grants
Tier 1 — Swiss foundations supporting research with project-bound cash grants (no equity). Active partners and historic supporters (since 2011).
Active
Historic
Public funding track record
Non-dilutive public funding (Swiss Innovation Agency Innosuisse + historic KTI). Cumulative cash & total-volume signal R&D maturity and validated research-partner network across 16+ years.
Licensing partners — IP licenses
Tier 2 — companies licensing brand IP for commercial use (royalty or fixed fee, no equity). Academic and public-research use stays free for citation; commercial use is private-capital licensing.
Science & market intelligence
The rehabilitation market is shifting from selling devices to proving recovery that survives daily life. The decisive layer is no longer the machine — it is the evidence that a movement is meaningful, not merely present. Motion is not recovery. This is the layer the science below builds toward.
Two trust anchors investors care about: does Joeri Gredig understand the science behind the products? and does Joeri Gredig understand the market they enter?
Closing the Substrate Gap in the JEPA Family
Working paper positioning Joeri Gredig's patents (EP 2 364 686 B1 granted 2015, PCT/EP2024/076634 drive-unit pending) inside the LeCun JEPA family. Introduces HCWM (Human-Centered World Model) as a structurally distinct construct, anchored in active-inference notation and instantiated as a 4-actuator EtherCAT testbed at 100 Hz hard-real-time. The prior art cited against these two families has been read in full — in none of the sixteen documents is the relative position of the limbs measured (applicant’s reading; European examination ongoing).
↓ Download paper (374 KB)Harvest Research Engine (HRE)
HRE is a continuous web-intelligence pipeline that tracks competitor filings, market signals, regulatory shifts, and scientific publications across each brand's vertical. Its feasibility was confirmed by an Innosuisse Innovationscheck (FHGR / Albert Weichselbraun, approved 2020). Operationalised on NVIDIA Jetson Orin Nano edge hardware (data_fetch_assistant.py).
The HCWM paper above is the flagship of a three-paper programme. Two companion papers extend the same patent-disclosed substrate — both pre-arXiv and available on request; their acknowledgements are pending contributor consent.
Counterfactual Latent Prediction (Predictor) — a world-model that asks not “did the patient move?” but “would this movement still hold under changed conditions?” Built on do-operator counterfactuals in latent space: the computational form of motion is not recovery.
The portfolio does not start from a blank page. It rests on validated aquatic-exercise and neuro-rehabilitation research, archived in a sovereign studies repository:
- Fenzl et al. — Release of ANP/BNP & fat oxidation in aquatic aerobic exercise (2012)
- Fenzl — Heart-rate variability for aquatic performance diagnostics (2013)
- Bansi et al. — Aquatic endurance training, multiple sclerosis (2013)
- Rezek — Randomised controlled trial, study protocol (2013)
- Rewald — Aquatic cycling (2018)
- Clijsen et al. — NIRS muscle-oxygenation measurement (2023)
- Frei, Schöch, Joos — Exergames in rehabilitation (2023)
- + contemporary reference: HIIT & protein turnover, Nature Communications (2026)
The field's leading authority describes — independently — the same axes this work is built on: intelligence as predict-then-realize, closing the loop between prediction and realization, grounded in embodiment.
- R. Riener (Ed.), AI to Eye — Between Code and Conscience, vdf Hochschulverlag Zürich, 2026 (ISBN 978-3-7281-4228-3)
- H. van der Kooij, …, R. Riener — “AI in therapeutic and assistive exoskeletons and exosuits: Influences on performance and autonomy”, Science Robotics 10(104), 2025 — DOI 10.1126/scirobotics.adt7329
Robert Riener — Professor of Sensory-Motor Systems, ETH Zürich; initiator of Cybathlon. Cited as independent reference, not as endorsement or affiliation.
How we know what we know
Three Innosuisse rejections, one ChatGPT launch in between, and the lesson that survives both: tools age in months — domain expertise grows in decades. The grail of the AI era isn't the best tool. It's the curator who knows which tool to point at which question.
Four questions every investor asks
Not the tool. Not the data. The curation — by an operator who lived three rejections, watched a language model obsolete his scientific edge in 19 days, and built the synthesis that survives both. That is what carries an investor across the gap between "interesting product" and "fundable thesis".
Readiness & market trend
Two views of one question — how ready is this portfolio, and is the market moving our way? The outside-in trend (what the field demands by 2030) and our own inside-out readiness scoring meet here, and feed the investment phases and the valuation.
The trend has moved toward our approach
Between May 2025 and August 2026 the trend line itself shifted — toward what craPos / HCWM has stood for: force-aware, safety-critical execution; reliability over hype; sovereignty; orchestration of intelligence layers. The third iteration of our Trend-Soll vs. IST gap-matrix (29 August 2026, twelve weekly issues read from 13 June on) reads 5 strong / ahead, 10 in progress, 0 critical gaps across 15 dimensions — three of them added since June, because the field itself named them: touch ladder (who may intervene, and how far), reviewability (what can be proven about one session) and, newest, the verification layer (what can be proven across many).
The sharper finding is not that the trend moved. It is that the value is migrating out of the actuation layer and into verification — the field now argues that the defensible position belongs to whoever can show, fast enough, that an action was the right one for that patient. On the machine side we can already put a number on it: 10.0 ms cycle time, 0.03 ms maximum jitter, 0 software faults over 6000 cycles. On the clinical side we say plainly what is missing: a record that joins one session to the next. That gap is why the newest dimension is also our weakest — and why we name it here rather than wait to be asked.
Cybersecurity — built in, not yet third-party provable
The sharpest 2026 demand (board-priority #1) is answered by design rather than deferred: security-by-design oriented to IEC 62304, hash-chained SHA-256 audit logging in an own certification service, and a documented device-hardening path (IEC 81001-5-1), mapped as "Next" in the passerelle. Stated plainly: that service currently sits behind our own single sign-on, so for a third party the evidence is asserted, not inspectable. Opening it to auditors is a task, not a research question — and EUDAMED (since 28 May 2026) and swissdamed (from 1 July 2026) still have to be assessed for which of our devices they bind.
A six-layer Neuroplasticity Operating System — a framework articulated in the field (notably by Zen Koh, June 2026) — describes the architecture rehabilitation needs. The portfolio already instantiates all six layers:
| Recovery-system layer | Our component | |
|---|---|---|
| 1 · Impairment map | HCWM / Predictor — reconstructs the hidden neuromotor state | v0 |
| 2 · Signal interface | craPos™ — force-aware robotic interface, 100 Hz, “runder Tritt” | built |
| 3 · Adaptive challenge | craPos gaits + Δ-parameters · ECP-UGC™ composer | built |
| 4 · Feedback engine | ECP-UGC™ game feedback + craPos live visualization | built |
| 5 · Dose engine | NAMRU (mobile/home) + ECP-UGC™ home scaling | concept |
| 6 · Translation layer | ascros™ medical pathway (Class IIb) + clinical Δ-outcomes | reg. |
Since that June instalment the same author has named five further frameworks, the most recent being the verification layer above — the six layers are where this began, not where the field stopped. The recovery loop is the defensible position — and that loop is the closed loop disclosed in EP 2 364 686 B1 / PCT EP2024/076634. Cited as independent observation of the field; no affiliation or endorsement implied.
Read the full mapping →Inside-out · readiness across the portfolio. The seven readiness levels (RRL·TRL·SRL·IRL·BRL·HRL·LRL, scale 0–9) per product — the same scores shown on each product card, laid side by side so the portfolio-wide pattern becomes visible.
| Product | RRL | TRL | SRL | IRL | BRL | HRL | LRL |
|---|---|---|---|---|---|---|---|
| ECP-UGC™ | 8 | 8 | 7 | 6 | 4 | 5 | 5 |
| craPos™ | 7 | 6 | 5 | 6 | 4 | 6 | 6 |
| Patello™ | 7 | 6 | 5 | 5 | 3 | 5 | 4 |
| ascros™ | 5 | 5 | 4 | 4 | 3 | 5 | 5 |
| craPos-dpcla™ | 5 | 5 | 4 | 4 | 3 | 5 | 4 |
| cranos™ | 6 | 4 | 3 | 3 | 2 | 4 | 4 |
| TRC™ | 6 | 4 | 4 | 3 | 2 | 4 | 3 |
Reading the columns: research & technology are the strength (RRL/TRL highest). Business Readiness (BRL) is the lowest column across every single product (2–4) — not a weakness to hide, but the precise thing a funding round buys. The technology is de-risked; market access is what capital accelerates. But the buyer is not missing — it is conditional: rehabilitation clinics tell us they would adopt ascros once its clinical hypotheses are confirmed. The gate is evidence, not willingness. That confirmation — proof that a gain survives daily life — is exactly what a funding round buys, and what unlocks procurement and reimbursement.
Place a bid
Non-binding offer. Pick an asset and phase, then name your amount. The model weights it by the phase you enter at — the earlier the phase, the more weight your money carries. Nothing is binding, and nothing is sent until you press the button. Joeri Gredig reviews each bid personally and replies with an AVPs offer — usually within two working days.
How this is computed
AVPs = offer × phase_multiplier × (1 + α × γ)
Share-valuation model v2.0. The phase multiplier is the inverse cumulative probability of
surviving the earlier phases; α is set per asset, γmedtech = 0.50.
What you actually receive
Your money enters as a convertible loan to Swissrehamed GmbH — you are not buying shares today. AVPs are the unit your loan converts by, not shares themselves. Until conversion you are a lender: no shareholder rights, no vote. Conversion is triggered by the next valuation round, by Swissrehamed’s conversion into an AG, or by an exit. The loan agreement governs.
Passerelles
Investment phases (6)
Investable entity today
Roadmap
Skeleton (live): read-only asset vitrine — products, passerelles, phases, entities.
Bidding (live now): non-binding bids → Joeri Gredig review → AVPs offer.