AK Venture Corp (AKVC) is a company that backs the stack where an agent’s claims can be checked: calibrated evidence scoring for what is still uncertain — candidates that PASS — and formal proof for what counts as Verified. AKVC Labs runs that verification platform. This is not LLM-as-proof.
The approach, and the verification platform at AKVC Labs.
Thesis
Production agents rest on probabilistic foundations. That is useful for proposing work. It is not a proof, and it is not a substitute for one.
AK Venture Corp (AKVC) is a company that backs teams who make the difference explicit. We look for decision models that generate continuous, calibrated evidence — PASS candidates with a known disposition to abstain when the support is thin — and for Lean and formal-verification kernels that say what is Verified: a machine-checked property, not a confident paragraph.
We look for companies building that stack, or products that sit on it, in settings where an uncheckable claim is a product defect — target domains include clinical decision support, autonomy, privacy-preserving systems, and authorization. AKVC Labs does that checking. The approach is how that work is read.
Why now
High-cost-of-error settings are adopting agents that cannot emit checkable claims. The missing company is a team shipping an evidence layer with a verification kernel. AK Venture Corp (AKVC) is a company that works in that category. AKVC Labs runs the verification platform.
Lean 4 and modern theorem-proving stacks are usable in product engineering, not only research labs: better automation, libraries, and AI-assisted proof tools that still defer to the kernel.
This category needs a reader who can take a spec, a kernel, and a go-to-market in the same sitting.
The work points toward clinical, autonomy, privacy, and authorization systems — places where trust is a product requirement, not a brand claim.
Portfolio
Flagship portfolio company: WorkProof — the trust layer between job completion and payment.
Verifiable work infrastructure
A completed work order is only a claim. WorkProof turns field evidence into a verifiable Proof Contract — so asset owners can accept work, release payment, and retain a defensible audit trail.
Policy decides. AI assists. Humans resolve uncertainty.
Proof Contract
WorkProof is a portfolio company, not an AKVC product. It fits the thesis because it treats completion as a claim, not a conclusion. Each required claim is checked against evidence, with explicit abstention when support is thin; capture carries provenance and anti-replay controls; and policy plus human review — not a model score — decide what is accepted. That is calibrated evidence under a verification kernel, not LLM-as-proof.
The trust layer between job completion and payment.
Framework
Evidence-scored assessment
Jev Quadrant is the framework AK Venture Corp (AKVC) publishes and uses to assess companies. Scores are evidence-based, reproducible, and calibrated.
Scope
Where the company looks. AKVC Labs runs the verification platform. We back teams building the stack, from calibrated evidence models and Lean kernels to domain products that sit on them.
We look for tooling that lets teams specify agent behaviour, emit machine-checkable properties, and re-check them as the system changes.
We look for companies building proof infrastructure for clinical decision-support, diagnosis assistance, and monitoring agents — designed so regulatory evidence can be assembled, not retrofitted.
We look for teams applying Lean 4 and dependent types to planning, decision policies, and interaction protocols — a kernel that decides what Verified means.
We look for formal methods for assistive robots around vulnerable people: motion, interaction, and fail-safe behaviour in messy care environments.
We look for tools that check privacy properties of agents handling advertising, insurance, or clinical data — mathematically stated guarantees, not policy copy.
We look for access policies that can be machine-checked for completeness and correctness — cloud security and zero-trust designs that fail closed.
We look for machine learning that accelerates proof synthesis and formalization — tools that help the kernel, and do not replace it.
We look for learning systems that teach formal methods with interactive, machine-checked feedback — growing the people this stack needs.
How we work
We read claims before we back a company. The full account is on the approach page.
Watch formal methods and agent research for technical turns that have product pull.
Calibrated evidence and, where it can exist, a formal proof inform what we read. A score is not a decision.
A score is about a claim. A kernel judgement is what Verified means. The written note is a person’s, with the limits stated.
Help on architecture, go-to-market, hiring, and the people who actually need a record.
Positioning
The company works on formal verification and calibrated evidence for agents — not a thin slice of a generalist brief.
We read proof infrastructure, not only decks. Research-heavy products are in scope.
The work points toward settings where an uncheckable agent is a defect: clinical, autonomy, privacy, authorization.
PASS is a candidate. Verified is a kernel judgement. We help teams keep those words honest.
Programmes
These programmes are AKVC Labs’ work on the verification platform. Register interest for any of them.
Get involved
Teams and researchers working on checkable agent claims. Questions about a programme can use Register interest on the programmes page.