Projects.
Krnali Labs is where I explore ideas around safe AI enablement and digital trust, through experiments, prototypes and writing.
It’s a place to build, test assumptions and share what I’m learning. If you’re working on similar questions, I’d be glad to compare ideas or try something together.
Projects
Questions we’re exploring.
How can agents act within clear boundaries? What can operational evidence tell us? How can digital interactions leave a record that others can verify? These projects explore those questions in practice, each with its own scope and stage of development.
Run agents with explicit authority.
Krnali SAR explores how AI agents can act in business workflows with access limited to the task and approval where it is needed. The aim is to keep people in control of what agents can access and change.
Move from an agent's proposal to an authorised action and a result you can check. Keep a clear record of what was requested, who authorised it, what ran and what happened, so teams can review outcomes and investigate when something goes wrong.
Why instructions are not controlStart with what the system can show.
Krnali Ops turns Kubernetes observations into findings, baselines and operational context. The observer stays read-only, inside the customer's environment.
Supported remediation has a separate actuator and an explicit approval boundary. The current action raises a Deployment's memory limit, with checks, verification and rollback. A person can grant revocable permission for a recurring, previously verified fix. Seeing a problem does not grant authority to change it.
Read the Ops updateLeave evidence that can be checked.
Krnali Verifiable Evidence (VE) is our SCITT implementation for verifiable credential workflows and AI agent transparency. It records signed statements in a transparency log and returns cryptographic receipts that can be checked independently.
Our development integrations retain evidence of credential issuance and verification, alongside agent outcomes and authority decisions. People can inspect the source observations, export the evidence and verify it offline. A receipt proves that a signed statement was recorded; it does not make the underlying claim true.
Explore Krnali VEConsulting services
Ways to work together.
Alongside the lab projects, I’m open to helping teams think through a specific question, test an idea or develop a proof of concept. These are some of the areas where my experience may be useful.
Find the work where AI can help.
We help organisations identify useful AI applications, assess readiness and turn a broad ambition into a practical adoption plan. The starting point is the work people do, the information they use and the decisions they remain responsible for.
- Use-case discovery and prioritisation
- Readiness and workflow assessment
- Pilot planning, success measures and team enablement
Connect AI to the way the business runs.
We design and build AI-assisted workflows around existing systems, data and people. That includes the integration itself and the boundaries around it: what an agent can access, when a person must approve, and how to recover when a step fails.
- Workflow and integration design
- Agent, tool and business-system connections
- Human approvals, evaluation and exception handling
Make the operating boundary explicit.
We help teams plan how AI will run, what it can change and how its behaviour will be observed. Identity, credentials, action permissions and operational evidence need to remain clear as experiments become working services.
- Deployment and access-boundary reviews
- Monitoring, evidence and incident workflows
- Controlled changes, verification and rollback planning
Turn a wallet capability into a useful process.
We work with organisations to explore where European Digital Identity wallets and verifiable credentials can support a real business decision. We develop the use case, map the trust relationships and build prototypes that make the assumptions testable.
- Remote verification and credential use-case discovery
- Holder journeys, consent and relying-party requirements
- Proofs of concept, integration design and acceptance testing
Put the rules where the work happens.
We help develop AI policies and governance that teams can apply in daily work. That means named owners, clear decision rights and review processes, with technical controls and evidence that support the policy.
- AI use policies and accountability models
- Approval, escalation and exception processes
- Evidence requirements and ongoing review
From agents to people
Trust crosses the boundary
of a single system.
Our EUDI messaging proof of concept explores remote identity verification inside a conversation, with the person choosing what to share from their wallet.
Exploring similar questions?
If you’re building something, testing an idea or working through a question around AI and digital trust, I’d be glad to compare notes. There may be something useful we can explore together.
Talk to us