Denzo Labs, a founder-led AI product company in London

Practical AIfor the peoplerunning the shift.

I build AI tools to find the right file, spot equipment problems and ask you before an agent takes the next step. I start with one task and test it with the people doing the work.

Start a pilotSee what works today
Built by Denis Lak. I’ve spent nearly 14 years looking after technology for restaurant and leisure groups.
Open-source tool publishedTwo working prototypes
A restaurant floor during service

Friday, 8pm.Where every tool gets judged.

What the AI is doingsample data
A preview of how these tools could help. Sample data, playing on a loop. Hover to pause.
June 2026Denzo Labs founded in London
23+sites rebuilt through one business transition
13food markets across 7 countries, in an earlier role
A chef at work in a busy kitchen

The pass.Where the work actually happens.

From my earlier roles, before I started Denzo Labs.

Tools for work that needs doing.

Here’s what I’ve been building for developers and the teams running busy venues. These demos use sample data; the labels show what’s ready and what I’m still working on.

For developersPublished · open source

agent-ping-cli

An agent asks you before it acts

Send up to three choices to your phone and return your decision to a waiting coding agent. Built for Claude Code and other coding agents; direct Claude API tool-use integration is in development.

Human approvalSample request
agent-ping: new decision request
Decision requestagent-ping
Prepare a consolidation plan?

Draft the proposed changes for review. No files will be moved or deleted.

Approve draftMore detailCancel

Return the approved choice to the waiting workflow.

    Useful when a workflow needs a person to approve the next step.

    Explore agent-ping on GitHub ↗
    What it does & what’s next

    Up to three choices per request, with expiry and replay handling, queueing, Node.js and PowerShell support. Published under the MIT licence.

    Next: connect it to Claude API tool use and develop shared policies and execution records for teams.

    For operations teamsWorking prototype

    Document intelligence

    Find the real copy of a file

    I built this to make sense of files scattered across a laptop and Google Drive. It compares their contents, finds exact copies and flags versions that need a decision. I’ve tested it on thousands of files. Dropbox is planned.

    Three storage systemsSample files
    Google Drive
    Supplier agreement.docxInventoriedHash a4e9…Kept as originalLinked in indexOriginal
    Laptop
    Supplier agreement (copy).docxInventoriedSame hash a4e9…Duplicate, file keptLinked, file keptExact match
    Dropbox planned
    Supplier agreement v2.docxInventoriedDifferent contentWhich version governs?Queued for reviewReview
    Unified indexReading three storage systems.Content establishes duplicates; names alone do not.One exact duplicate. One uncertain version.One index across all three. Nothing moved or deleted.

    Link every copy into one index and keep files pending review.

      Useful when shared folders contain duplicates and competing versions.

      Explore the local scanner ↗Discuss a document pilot ↗
      What it does & what’s next

      It groups files by their contents, brings overlapping folder lists together and keeps a dated record of changes. If it can’t tell which version to use, it leaves that decision to you.

      Moving files between services and carrying their permissions across still need more work.

      For venue teamsDevice prototype · restaurant concept

      Connected operations

      Catch an equipment issue before service

      I’ve connected a chat assistant to selected devices through Home Assistant. Next, I want to bring that into a restaurant: spot a fridge warming up before service and prepare a maintenance note for the manager to check.

      Service readinessSample scenario
      Site 04 / lunch service
      Bar fridgeTemperature history, door state
      Within limitOver limitOver limitOver limit
      Illustrative temperature trend rises above a site-configured alert limit. Site alert limit Earlier readings Now
      Pre-service check in progressTemperature above the site's alert limitRising with the door closed. Cause unconfirmed.Exception ready for manager review
      Sample readingsDoor: closedTrend: rising
      Check equipment contextRead recent history before surfacing an exception.Flag the exceptionOne useful alert, with the affected site and asset.Prepare a handoverInclude the trend and door state; avoid guessing a cause.Maintenance draft readySite 04, bar fridge, readings attached. Manager review needed.

      Prepare a maintenance draft for manager review.

        For teams who need to know what’s wrong, where it is and who needs to act.

        Discuss an equipment pilot ↗
        What I’ve tested & what’s next

        The current scope is 12 lights, with automatic exposure disabled. OAuth access and light control were tested in the ChatGPT desktop app.

        Next: resolve web/mobile access, extend device integrations and test a repeatable pilot deployment. Voice interaction, managed approvals and execution records are product directions.

        The restaurant walkthrough illustrates a planned sensor and maintenance workflow, not a deployed restaurant system.

        The demos play on their own. Hover to pause, or pick a step to take a closer look.

        How a pilot works.

        Bring me one task that takes too much time or keeps going wrong. We’ll agree what to test, what I can access, the cost and the timing before I start.

        What you get

        Something you can try on your own work, a conversation about the results, and clear instructions for whoever will look after it.

        What we agree first

        What we’re solving, which files and systems I can use, and how we’ll know it helped. You’ll see the plan before we begin.

        1
        You bring one problem.Folders full of duplicate files, an agent that needs your approval, or equipment nobody has time to check.
        2
        We scope it together.I watch the work being done, agree what data and systems I can access, and define what a useful result looks like.
        3
        I build and test on your data.You judge the output on your own files and numbers, not a demo.
        4
        You get a handover.An owner, documentation and clear boundaries, so it keeps working without me.
        Staff talking at a restaurant bar

        Start with the actual shift.Watch the service, talk to the people doing the work, find out what breaks when it gets busy.

        How I like to work

        AI drafts, people decide.Let the tools gather information and prepare a draft. Keep the decision with the person responsible.
        Test on real data.Judge the output on your own files and numbers.
        Agree what I can access.Start with the files and systems we’ve agreed on.
        Say when I’m unsure.If something needs checking, I’ll tell you.
        Make it easy to look after.Leave clear instructions and someone who knows how it works.
        Say what isn't finished.Be clear about what works today and what I’m still building.
        Illustrated tech avatar of Denis Lak
        Denis LakFounder · London

        About

        I'm Denis. I've run the tech you're trying to fix.

        After nearly 14 years working full-time in technology, I started building seriously with AI tools in December 2025. In June 2026 I founded Denzo Labs to turn that work into practical products for operational businesses.

        Since then, I've published an open-source agent tool, built working prototypes and started working with early clients. I’m working on tools that help people approve agent actions, sort out their files and keep an eye on equipment. The approval tool works with Claude Code today; I’m building the direct Claude API integration next.

        LinkedInGitHub

        Background

        1. Early career

          Hands-on IT support, systems and day-to-day operations in hospitality and leisure.

        2. IT management

          2017–2020

          Managing technology across hospitality businesses, including international operations.

        3. Technology leadership

          2020–2025

          Head of IT and Head of Technology roles across restaurant groups and international food markets.

        4. Delivery & direction

          2026–present

          Senior delivery contracts, then Tech Director alongside founding Denzo Labs.

        EducationInformatics and Econometrics
        Wrocław University of Economics and Business

        What I’m working on in my day job

        Alongside Denzo Labs, I'm Tech Director at a London restaurant group. These are projects I’m working on with the team there.

        Agents that only speak up when it matters.
        Restaurant group, London Team project
        ProblemReports nobody opens, and issues spotted a week too late.
        DirectionScheduled checks of site data such as forecasts, kitchen speed, delivery availability and bookings against staffing. Surface useful exceptions in team chat, with causes stated only when supported by evidence.
        GoalHelp managers identify important exceptions without adding reporting overhead.
        From a stack of spreadsheets to one menu source.
        Restaurant group, London Team project
        ProblemDozens of menu variants rebuilt by hand twice a year. Prices, copy and artwork drifted apart.
        DirectionA shared menu database with separate content and pricing responsibilities, an approval workflow, and consistent print and web outputs. A read-only AI connector could support menu queries and sales analysis.
        GoalReduce version errors and make menu changes easier to manage and analyse.
        Rolling AI out to a whole head office.
        Restaurant group, London Team project
        ProblemKeen people, no shared rules. Data heading for the wrong tools, five tools doing the same job.
        DirectionRole-relevant training, a map of team use cases, connector approvals, access rules and a concise AI policy. Adoption should be assessed alongside data access and software costs.
        GoalSupport useful adoption with clear data and cost controls.
        First-line IT help, inside the team chat.
        Restaurant group, London In design
        ProblemVenue teams raise tickets for two-minute fixes nobody showed them.
        What I'm doingA proposed helpdesk assistant using approved guidance and past ticket resolutions to support staff, then escalate unresolved issues with relevant context.
        AimFewer tickets, quicker fixes mid-shift, a tech team freed up for bigger work.

        These are projects in progress. I’ll share results when there’s something measured to show.

        Start a pilot

        Bring one problem worth solving.

        Tell me what keeps going wrong or taking too long, and which tools you use. I’ll let you know if I can help and what I’d suggest trying first.

        Email DenisFounder-led, London