Selected work

A research and outreach workflow that finds companies matching defined criteria, gathers public evidence and prepares personalised emails. Drafts are checked and queued for approval, with outreach status and replies tracked in the same place.

An email assistant for enquiries about high-value products. It combines product documents and earlier replies to prepare multilingual drafts, which staff review, edit and send from their existing inbox.

An inbox assistant that brings customer records and previous conversations into the drafting process. It checks earlier discussions and prepares a reply in the normal email workflow for a person to review.

A shared operating system for leads, clients, delivery work, agreements and invoicing. It connects customer and project information with delivery evidence, reducing the need to move records between separate tools.

An internal signing workflow that prepares documents, sets signer roles and order, and tracks progress through secure links. Reminders, signed documents and the audit trail return to the system where the work began.

A directory platform that imports service-provider records, removes duplicates and enriches company information. It creates category and location pages from structured data, making recurring record and page maintenance easier to manage.

A creative production tool that turns existing images, logos and brand materials into reusable video compositions and variations. Repeated production steps are standardised while creative choices remain editable.

A design tool that uses existing brand assets, colours, typography and references to develop initial concepts and variations. A designer reviews the options and retains control over selection and final refinement.

An archive workflow for preserving useful AI-generated material with context. It prepares metadata, checks for duplicates and routes each candidate through human editorial review before publication.

A research pipeline that processes an existing company list and collects selected public company, financial and contact information. Findings are structured into a sourced dataset, with missing or uncertain information flagged for review.

A multi-agent research platform that generates candidate strategies and evaluates them separately against withheld historical data. The work explores repeatable evaluation and evidence-based selection; it does not establish investment performance or future returns.

A research platform that collects Nordic company data, classifies companies and surfaces funding and market signals. It prioritises potential opportunities against defined investment criteria for a person to assess.

A search system that queries independent indexes, removes overlapping results and combines rankings locally. It explores how to give a team more control over its source selection without manually repeating the same search.

A dictation tool started with a global keyboard shortcut. Speech is transcribed on the user’s device and made available in the current workflow, keeping recordings and transcripts local unless the user chooses to share them.

A browser automation experiment that turns written tasks into actions across website interfaces. It inspects page structure to choose controls and navigate multi-step flows rather than depending only on fixed click paths.

An interface for supervising locally hosted coding agents from desktop or mobile. Repositories, credentials and conversation history stay with the local working environment while the user directs work remotely.

A hosted platform that provisions a separate AI-assistant environment for each user. Persistent memory, messaging integrations, billing and lifecycle management are brought together to make assistant operations repeatable.

An operating-environment experiment where an agent describes system configuration as reviewable state. Proposed changes are tested before activation, with rollback available when a configuration fails or is no longer needed.

An execution-environment experiment that runs agents in restricted containers and routes authenticated requests through a separate credential proxy. It explores keeping raw secrets outside the agent environment; security properties require project-specific validation.

A development agent that receives requests through messaging and works in a persistent cloud workspace. It uses coding tools to implement and test changes, producing work for human review before acceptance or release.

A development workflow in which a lead agent plans and assigns tasks to worker agents. Implementation and testing can run in parallel, while completed work goes through an explicit review and integration process.

A communication-layer experiment that validates structured messages between agents and separates control fields from untrusted text. The approach makes message contracts reviewable and explores limiting the spread of prompt-injection content.

A prototype that reads an equipment LED through a camera and converts its state into machine-readable information. It demonstrates a way to connect an existing device to monitoring and alerts without modifying the equipment itself.

A workflow that turns written part requirements into parametric CAD models, checks defined printability constraints and prepares models for slicing. Final manufacturing decisions and printing remain under human review.