Know if a model transfers safely before you deploy. Structural transfer audit, cross-architecture compatibility check, tamper-proof certificates — delivered as an API.
“Whether it's a compressed vision model or a large language model — the failure starts in the structure. That's what we measure.”
21 open-source plugins
NVIDIA, AMD, MLflow, Hugging Face, vLLM, Triton, TensorRT, Kubeflow, Argo, BentoML, ClearML, more. Every pip install is a distribution channel.
Every pill links to its repo on github.com/LeeSandberg — MPL-2.0 licensed.
Every team shipping ML faces the same question: did quantization, fine-tuning, or distillation break something? Today validation is manual, ad-hoc, and invisible to auditors. As regulation tightens (EU AI Act, FDA, NIST AI RMF), “trust me, it works” stops being acceptable. There is no standard infrastructure for proving a transformed model transfers safely.
Transfer Oracle structurally audits ML model transformations and produces tamper-proof, HMAC-signed certificates that prove a model transfers safely. Live in production.
Independent structural probes verify model integrity beyond accuracy. Catches per-class blind spots that aggregate metrics hide.
Validate behavioural agreement between two models (large/small, GPU/edge, before/after distillation) on the same inputs. Just shipped.
HMAC-signed audit reports with hash-chained integrity. Drop straight into compliance evidence packages.
22 open-source MPL-2.0 plugins covering NVIDIA, AMD, MLflow, Hugging Face, vLLM, Triton, TensorRT, Kubeflow, Argo, BentoML, ClearML, and more.
White-label multi-tenant architecture. Operators retain 75–85% of revenue; we license platform, brand, infrastructure.
Output maps directly to EU AI Act Article 9, NIST AI RMF, and FDA SaMD evidence requirements.
What's built
Live API, live billing, live customers, live operator instances.
Pricing & revenue model
Path to €5M ARR: 200 enterprise customers at €25K ACV.
Market opportunity
MLOps market 2030
$0.0B$2.3B → $16.6B, 40% CAGR. Grand View Research.
Model validation 2033
$0.0B$1.8B → $6.5B, 17% CAGR. SNS Insider.
AI governance 2036
$0B$2.2B → $11B, 15.8% CAGR. Future Market Insights.
Combined MLOps + validation + governance
Model validation + compliance for production enterprises
Competitive landscape
Adjacent companies focus on monitoring and observability. Transfer Oracle occupies a distinct position: structural audit of model transformations with tamper-proof certification — a requirement under emerging regulation.
Acquired by Cisco (2024)
Acquired by Apple (2025)
$131M raised — Series C, Feb 2025
$63M raised
Regulatory catalyst
High-risk system requirements become mandatory August 2, 2026. Conformity assessments take 8–16 weeks. Notified bodies already backlogged into Q2 2026.
1,250+ AI/ML medical devices authorized. Each model change requires re-validation. Transfer Oracle automates the evidence package.
Referenced by FTC, SEC, and CFPB in enforcement actions. Audit-trail requirements align with our certificates by design.
Defensibility
Hash-chained audit certificate (filed). Cross-architecture compatibility methods, chain-contract format, bidirectional plugin adapter, and chat-driven chain generation queued for filing in 2026.
Every audit produces a signed, verifiable certificate. We become the trusted third party for model deployment decisions across the regulatory regimes that demand evidence.
22 open-source plugins embedded in NVIDIA, AMD, MLflow, Hugging Face, and orchestration stacks. Every pip install is a potential API customer.
White-label brand multiplication: each operator targets a vertical (ML, AI safety, future verticals) on the same infrastructure. Capital-light scaling.
Plugins designed to flow both directions — host tools call our audit AND our marketplace chains run inside host tools. Network effects: new chain reaches every runtime; new runtime gains every chain.
Chat-driven chain composer at /growt/studio with one-source-multi-output emission (MCP, API, CLI, Python module). Marketplace for publishing and discovering audit chains.
Take everything we've built and let customers compose it. /growt/studio — a chat-driven editor that emits one chain source to five runnable artifacts (MCP, API, CLI, Python module, LangChain LCEL). Marketplace — every chain published is instantly callable from inside MLflow, Hugging Face, vLLM, NeMo, and 18 other host runtimes via our bidirectional plugins.
Network effects: every new chain reaches every runtime; every new runtime gains every chain. n8n / Zapier compounding applied to AI auditing.
The round
We're raising €500K to take Transfer Oracle from live API to enterprise GTM ahead of the EU AI Act deadline. Looking for investors who understand infrastructure software and the regulatory wave ahead.
Founder: Lee Sandberg
Company: Droidtech 42 AI Labs AB, Org.nr 559534-0745
Location: Stockholm, Sweden
Stage: Pre-seed