Skip to content
Sign in
SWC 2026 FinalistGrand Final: Lunch Block

Neurath AI

Uncertainty-quantification infrastructure that tells critical-AI operators when their models are unreliable and which data would fix it.

United States, North AmericaAI & software, AI model reliability and robustness
Stage
UndeterminedAI at the core
Product
PrototypeNone known
Funding raised
Undisclosedno funding
Founded
20251 years ago
Founders
—
Global score
3.5Grade DRank 111 of 151

Overview

Reliability infrastructure for critical AI (robotics, defense, medicine, industry): measures the predictive stability of models, maps their blind spots and targets the data needed to fix them.

Traction

None known

Trireme SDK and Gauloi platform described on the website, but no public customer, deployment or contract found.

Market

AI trust, risk and security (AI TRiSM)

As AI moves into robotics, defence, medicine and industry, buyers must prove to regulators and customers that models behave reliably outside training conditions. The EU AI Act, medical-device rules and defence assurance frameworks are converting this into a budget line (AI TRiSM). Buyers are ML platform leads and safety/compliance officers; sales cycles in defence and healthcare are long but contracts are sticky.

Addressable market

2025
$2.9B
2030
$7.4B
+21.6% a year

AI trust, risk and security (AI TRiSM): global market from $2.9bn in 2025 to $7.44bn in 2030 (CAGR 21.6%).

Customers

Defence primes and drone makers, robotics companies, AI medical-device makers, industrial AI teams

Who signs the cheque

Head of ML / safety assurance; compliance and regulatory affairs; defence programme offices

Geographies

  • United States
  • Europe (regulated buyers)

Drivers

  • AI regulation and assurance requirements
  • Deployment of AI in safety-critical systems
  • Cost of collecting the wrong training data

Headwinds

  • Long defence/healthcare sales cycles
  • Competition from MLOps and AI TRiSM platforms
  • Opaque company (no team or customers published)

Global AI TRiSM market from $2.9bn in 2025 (The Business Research Company) to $7.44bn in 2030 (Grand View, CAGR 21.6%).

Scalability 8/10. Software infrastructure (SDK) replicable worldwide; constraint: long sales cycles in defense and healthcare.

Market source: gii.tw

Technology

Category

ML infrastructure (AI assurance)

AI at the coreMachine learning / predictive

Maturity

  1. R&D
  2. Prototype
  3. Pilot
  4. On the market
  5. Scaling

prototype

How it works

Uncertainty quantification measures the predictive stability of models, maps blind spots and targets the data needed to fix them. Delivered as the Trireme SDK and the Gauloi Kubernetes platform, on-premises or in a VPC.

Moat

Potentially moderate: UQ methods are published research, but on-prem deployment with SOC 2, ISO 27001, NIST 800-171 and CMMC compliance is a real barrier for defence buyers.

Differentiation

Targets data acquisition (which data to collect) rather than only monitoring, and is built for air-gapped/regulated environments.

Integration points

  • MLOps pipelines
  • Model registries
  • Data-labelling and collection workflows
  • Medical-device QMS post-market surveillance

Depends on

  • Access to customers' models and data
  • Defence compliance (CMMC)
  • Kubernetes infrastructure at the customer

Use of AI

Uncertainty quantification that measures the reliability of AI models and maps their blind spots.

Product in detail

Uncertainty quantification. Trireme SDK and Gauloi Kubernetes platform deployed on-premises/VPC (SOC 2, ISO 27001, NIST 800-171, CMMC compliance).

Business

Model
Enterprise software licence (SDK + platform), likely annual on-prem licences (not published).
Revenue logic
Per deployed model or per platform instance; value = avoided failures and reduced data collection.
Go-to-market
Direct enterprise sales to defence, robotics and medtech; pilots with integrators.

Next milestones

  1. Publish team and first design partners
  2. A defence or medtech pilot with measurable blind-spot reduction
  3. Non-dilutive defence funding (SBIR/DIANA)

Key risks

  • No traction
  • Opacity
  • Long sales cycles in defense and healthcare

Opportunities

4 found in the network

Concrete moves for Neurath AI inside the network: who to sell to, sell through, build with or raise from, each with the reasoning, the evidence and the first step.

  • Customers 2
  • Co-development 1
  • Investors 1

Customers2

Who could buy, or sell through

Confidence bounds for GPS-free positioning

OmniLoc's AI computes absolute position from camera-to-satellite matching; drone OEMs and defence users need to know when that estimate is unreliable. Neurath's uncertainty quantification can deliver calibrated confidence and identify terrain where more imagery is needed.

35%confidence

Evidence from the records

  • “AI matching of camera views with satellite imagery to locate drones and vehicles without GPS.”
  • “Uncertainty quantification that measures the reliability of AI models and maps their blind spots.”
If it works
A safety-assurance feature in OmniLoc's OEM pitches and Neurath's first defence-adjacent reference.
First step
Offer OmniLoc a free blind-spot mapping on one validation dataset.

Post-market reliability monitoring for a CE-marked AI device

Legit.Health's CE-marked dermatology AI must monitor performance across skin types and image conditions; Neurath maps blind spots and the data needed to fix them, aligned with MDR post-market surveillance.

25%confidence

Evidence from the records

  • “CE-marked medical device (ISO 13485), with partners listed such as NHS, SERMAS, SESPA, Sanitas, Adeslas, Novartis and Almirall.”
  • “Reliability infrastructure for critical AI (robotics, defense, medicine, industry)”
If it works
Regulatory-grade reliability evidence for Legit.Health; a medtech reference for Neurath.
First step
Send Legit.Health's regulatory team a proposal for a retrospective blind-spot analysis.

Co-development1

An integration or joint product

Joint AI assurance stack for defence

Hardshell protects training data against poisoning and leaks; Neurath measures model reliability and targets new data. Together they cover data integrity and model robustness, the two pillars of defence AI assurance, and both target defence buyers.

30%confidence

Evidence from the records

  • “Hardshell protects sensitive data used to train AI/ML systems against poisoning and leaks.”
  • “Uncertainty quantification: Trireme SDK and Gauloi Kubernetes platform deployed on-premises/VPC (SOC 2, ISO 27001, NIST 800-171, CMMC compliance).”
If it works
A bundled offer for defence SBIR/primes with higher win probability than either alone.
First step
Schedule a call with Hardshell's founders to scope a joint SBIR topic response.

Investors1

A fund whose thesis fits

Dual-use accelerator for AI assurance

NATO DIANA funds pre-seed/seed dual-use technology across Europe and North America and backed network member Astrolight, a warm source of application advice. AI reliability for defence systems is a core DIANA challenge area.

20%confidence

Evidence from the records

  • “Critical defense theme”
  • “Relevant theme”
If it works
Non-dilutive funding plus access to NATO test centres and end users.
First step
Ask Astrolight about their DIANA application and monitor the next challenge call.

Analyst reading of 21 Nov 2026, from the public record of the network. Method: analyst-ai.

Investors that fit

Sarah Kunst's pre-seed firm writing up to US$1M cheques, mostly to underrepresented founders outside Silicon Valley, in AI, cybersecurity and fintech.

US AI reliability/robustness platform (prototype): AI security-adjacent at pre-seed. Founder profile vs Cleo criteria unknown.

35%confidence

Evidence from the records

  • “AI model reliability and robustness”
  • “Prototype”
Intro path
Hardshell (Cleo portfolio)
First step
Apply to Cleo's cyber accelerator via Hardshell.

US investor with no profile in our data; joined Hardshell's $1.1M pre-seed in AI training-data security.

US AI reliability and robustness — AI trust and security.

30%confidence

Evidence from the records

  • “Hardshell”
  • “AI model reliability and robustness”
  • “Trireme SDK and Gauloi platform described on the website, but no public customer, deployment or contract found.”
Intro path
Founder-to-founder intro via Hardshell (network portfolio company of MoJo Ventures).
First step
Ask Hardshell for an intro.

US investor with no profile in our data; joined Hardshell's $1.1M pre-seed.

US AI robustness.

30%confidence

Evidence from the records

  • “Hardshell”
  • “AI model reliability and robustness”
  • “Trireme SDK and Gauloi platform described on the website, but no public customer, deployment or contract found.”
Intro path
Founder-to-founder intro via Hardshell (network portfolio company of Not Yet Ventures).
First step
Ask Hardshell for an intro.

US federal department funding dual-use startups through ESTCP and xTechSBIR, as received by waste-to-energy company Enexor.

Reliability infrastructure for critical AI including defence: relevant topic area, no traction found.

25%confidence

Evidence from the records

  • “Reliability infrastructure for critical AI (robotics, defense, medicine, industry)”
  • “Prototype”

US investor in Hardshell's US$1.1M pre-seed (AI training-data security); thesis and ticket are undocumented.

US AI model reliability and robustness platform: AI-trust infrastructure adjacent to Hardshell.

25%confidence

Evidence from the records

  • “AI model reliability and robustness”
  • “United States”
Intro path
Hardshell (Black Prism portfolio)
First step
Ask Hardshell for an intro.

US space agency funding space technology through contracts and SBIR grants — backer of TRL11, G-Space and Opal AI.

US AI reliability — autonomy assurance topics.

25%confidence

Evidence from the records

  • “"United States"”
  • “AI model reliability and robustness”
  • “Trireme SDK and Gauloi platform described on the website, but no public customer, deployment or contract found.”
Intro path
Founder-to-founder intro via G-Space, Inc, Opal AI, TRL11 (network portfolio company of NASA).
First step
Map SBIR topics.

US investor with an empty profile; participated in Hardshell's $1.1M pre-seed (AI training-data security, 2026).

Reliability infrastructure for critical AI: AI-assurance adjacent to Hardshell's data security.

25%confidence

Evidence from the records

  • “AI model reliability and robustness”
  • “United States”
Intro path
Hardshell founders.
First step
Ask Hardshell about VTC's process.

Investment arm linked to Protocol Labs; the only recorded data point is the oversubscribed $7.5M seed of brain-interface company Neurosoft.

Reliability infrastructure for critical AI, including medicine; a frontier-AI angle that may fit a tech-forward investor, but no evidence of PL Capital AI deals.

15%confidence

Evidence from the records

  • “Reliability infrastructure for critical AI (robotics, defense, medicine, industry)”
  • “Trireme SDK and Gauloi platform described on the website”
First step
Confirm PL Capital's AI mandate before any outreach.

Investment analysis

3.5D

Grade D

Rank 111 of 151

Disruption6.0/10
Business model1.0/10
Market5.0/10
Risk(lower is better)9.0/10

Global score = mean of disruption, model, market and (11 − risk). Last updated 21 Nov 2026.

Not investable as it stands.

Disruption

6/10

Reliability and blind spots of critical AI models: interesting approach.

Business model

1/10

No customers or contracts.

Market

5/10

TAM AI trust, risk and security (AI TRiSM) ~$2.91bn (base 4), strong scalability 8/10 (+1), no proven international expansion (0) → 5/10.

Risk

9/10

Lower is better

Very high risk: unknown team, no data.

Strengths

  • Relevant theme
  • SDK described

Risks

  • No traction
  • Opacity

Funding

Undisclosed, no funding

Amount used in the study. No public funding round found.

Note. no public funding round found

verified: website with no team or customers; no founder, funding round or SWC link found through search.

Team

No founders listed yet.

Sources

Identification confidence
medium
Founding year: 2025 (low confidence)
First website archived in 2025, proxy (year the website existed, no confirmed founding date) web.archive.org
  1. 1neurath.ai/

From the open study Startup World Cup 2026 by TIBOK, based on public sources, verified in November 2026. Scores are an analyst's judgment on an explicit grid, not the jury's.

Wall

Wall

No posts yet