Real Projects Real Numbers

Real Engineering Problems.
Measured Results.

Here are three standout projects from our work across various clients. These aren’t three projects we completed. They’re three expensive engineering problems: the kind that quietly eat engineering time, delay releases, or block a launch. We fixed them, and we kept the before-and-after numbers to prove it.
We understand you are looking for real client names. We safeguard client names in success stories under strict NDAs. Delighted customers are ready to provide references upon request.

Embedded Systems  ·  IoT  ·  Yocto

They Were Losing 4 to 6 Hours Every Time They Built Firmware

Their engineers, at a global audio device manufacturer, were spending up to six hours building each firmware image by hand. Every build and related activities release was done manually. No proper version control, no visibility into what was happening, and a product portfolio that kept growing. We rebuilt the whole process. The same build now runs in 45 minutes on its own, while the team works on something else.

  • 8x Faster builds. Was 4 to 6 hours. Now 45 minutes.
  • Zero Manual steps needed to run a build
  • On-demand Releases. They used to ship monthly and on a fixed schedule.
  • Near zero Error rate. Previously high on every release.

Background

A global manufacturer of audio devices came to us with a build problem they’d been living with for a while. Every new firmware release meant an engineer sitting down and running the Yocto build by hand. One image took between four and six hours. That was just the build itself, before any testing or release work.

The team had no central build machine & process and they were planning to migrate to cloud, and no record of what had changed between builds. If something broke, figuring out why took time they didn’t have. And their product line was growing, planning to achieve is a compliance certification like ISO 27001, which meant the problem was only going to get worse.

What They Were Dealing With

  • No traceability of binary artifacts and standard SCM and Release engineering practies like no record of what changed between builds.
  • Every build was done by hand. Errors crept in regularly and were hard to trace.
  • Leadership had no visibility into build status. Release managers were chasing engineers for updates.
  • Six-hour builds meant the engineering team was stuck waiting, not shipping. With a growing product portfolio, this wasn’t sustainable.

What We Did

  1. Moved the codebase to GitHub and set up proper version control, branch policies, and team workflows. Everything the team should’ve had to begin with.
  2. Built automated Yocto build pipelines using GitHub Actions. Every code push now triggers a full image build automatically. Nobody has to start it.
  3. Added proper logging and monitoring so engineers and leadership can see exactly what’s happening during a build and debug failures quickly when they occur.
  4. We identified the optimal Azure configuration to host the build process, delivering a smooth cloud migration while keeping infrastructure costs to a minimum.
  • Set up cloud artifact storage so every build output is versioned and accessible. Release management across multiple product lines became straightforward.

    • Before and After

    MetricBeforeAfter
    Build duration4 to 6 hours45 minutes
    Process typeManual, done by hand each timeFully automated
    Error rateHigh, on every releaseNear zero
    Release frequencyMonthly, fixed scheduleOn-demand, whenever they need
    Dedicated build infraLocal laptopBuilds happening in Azure Cloud, Secure.
    Build traceabilityNoneFull, versioned cloud storage

    How we measured it: build duration compared directly, same hardware target, before and after the pipeline went live. All those builds were able to execute via Github Action UI fully automated

    Business Impact

    At 20 builds a month, that's roughly 20 × 5 hours of engineering time returned to feature work instead of babysitting a build. This is huge saving, just run TuskerGain for a defensible estimate.

    What the client said

    Stonetusker helped us move from a manual Yocto build process to a fully automated, cloud-based pipeline. Our engineering team now releases faster, with complete visibility and far less manual effort.

    Vice President of Engineering Global Audio Device Manufacturer

    Running a similar build process or manual or slow Embedded Linux software delivery pipeline? Talk to Us About Embedded DevOps See our Embedded DevOps approach
    Cybersecurity  ·  SaaS  ·  AWS & Azure

    They Were Burning Weeks on Compliance Reviews and Marketplace Admin

    A growing network security and fraud-mitigation company was doing almost everything by hand: deployments, SOC compliance reviews, and AWS and Azure marketplace admin. We automated the pipeline, the compliance workflow, and the marketplace registration. They now ship features 50% more often and spend 75% less time on marketplace admin.

    • 50% More deployments. Features get to customers faster now.
    • 60% Faster SOC compliance updates. Used to take weeks.
    • 75% Less manual work for AWS and Azure marketplace onboarding
    • Zero Downtime during database updates since automation went live

    Background

    A fast-growing company building network security and fraud mitigation products needed to ship software faster, keep up with changing SOC compliance requirements, and get their product properly listed in AWS and Azure marketplaces. They were doing most of this by hand, and it was eating up time their small team didn’t have.

    What They Were Dealing With

    • Deployments were slow and done manually. Errors came up regularly and delayed releases for customers.
    • SOC2 compliance updates were taking weeks each time. The requirements kept changing and the review process was entirely manual.
    • Listing and maintaining their product in both AWS and Azure required a lot of manual work that had to be repeated every time something changed.
    • Database updates during deployments were done by hand, which meant real downtime risk on every release.
    • Their app was built with Golang, which needed someone who knew how to automate Golang deployments reliably without creating new problems.

    What We Did

    1. Built a full CI/CD pipeline for the Golang application, covering build, test, and deploy. Deployment time came down by 40% and manual errors stopped happening.
    2. Automated the SOC compliance update process with full audit trails built in. What used to take weeks now takes days, and the team doesn’t have to babysit it.
    3. Scripted and standardised the AWS and Azure marketplace registration and update process. The scripting effort itself dropped by 70% vs. the 75% figure above, same measurement.
    4. Automated database migrations so they happen during deployments without any service interruption. No more planned downtime windows for a database update.

    Before and After

    MetricBeforeAfter
    Deployment frequencySlow, manual, irregular50% more frequent
    SOC compliance cycleSeveral weeks per update60% faster, days not weeks
    Marketplace onboardingHigh manual effort, slow turnaround75% less manual work
    Database update downtimeReal risk on every deployZero downtime

    How we measured it: deployment frequency from release history, compliance cycle time from ticket/audit-trail timestamps. Also, single click migration of their DB also saved a lot of time.

    Business Impact

    Features reach customers roughly 50% more often, directly shortening time-to-value for their own customers, and compliance stopped being a bottleneck on release velocity.

    What the client said

    Stonetusker has been helping my company with CI/CD for our AWS software workflows and has really excelled at improving our Git, GitHub, security scanning, issue tracking, and product testing capability and velocity.

    Chief Executive Officer Network Security Solutions Product Company

    Dealing with compliance overhead or marketplace complexity? Talk to Us About Platform Engineering See our Platform Engineering approach
    AI & MLOps  ·  Healthcare  ·  HIPAA & GDPR

    Their Cancer-Detection AI Was Ready. Compliance and Deployment Weren’t.

    A US-based AI cancer-detection startup was ready to launch, technically. Model training and deployment were entirely manual across four disconnected tools, with no rollback safety net, and HIPAA and GDPR compliance stood between them and the market. We built the pipeline and the compliance posture they needed. Model updates now deploy 60% faster, production has had zero downtime, and they launched fully compliant.

    • 60% Faster AI model deployment. Faster iterations mean faster accuracy improvements.
    • Zero Production downtime since automated deployments went live
    • Full HIPAA and GDPR compliance achieved before market launch
    • One pipeline Across Python, TensorFlow, PostgreSQL and ReactJS. All coordinated.

    Background

    A US-based startup was building AI-powered cancer detection software using medical imaging. The technology was promising, but their engineering setup hadn’t kept up with the pace of development. Model training and deployment were both done manually, their stack was spread across four different frameworks that didn’t talk to each other cleanly, and they couldn’t launch until they were fully HIPAA and GDPR compliant.

    They needed all three problems solved before they could go to market. The clock was running.

    What They Were Dealing With

    • No MLOps pipeline. Model training, testing, and deployment were all manual. Every update to the model required manual work to get it into production.
    • Deployments to development and production were done by hand. There was no automated promotion process, no approval gates, and no rollback capability if something went wrong.
    • Their tech stack used Python, PostgreSQL, TensorFlow, and ReactJS. These needed to be coordinated in a single deployment workflow, which they weren’t.
    • HIPAA and GDPR compliance wasn’t just a nice-to-have. They couldn’t launch without it. Patient data handling, audit logging, and access controls all needed to pass regulatory review.

    What We Did

    1. Built a fully automated MLOps pipeline. Model training, validation, testing, and deployment are all automated and version-controlled. The team no longer pushes models to production by hand.
    2. Set up GitHub Actions workflows for both development and production environments, with proper approval gates and rollback built in. Promoting a model from dev to production is now a controlled, repeatable process.
    3. Unified the deployment pipeline across all four stack components. Python backend, TensorFlow models, PostgreSQL migrations, and the ReactJS frontend all deploy together in a single coordinated workflow.
    4. Worked through the HIPAA and GDPR compliance process with them. Audit logging, data encryption, access controls, and all the required policy documentation were completed and signed off before launch.
    5. Stayed involved after handover to support the team through the actual launch period and the compliance questions that came up along the way.

    Before and After

    MetricBeforeAfter
    AI model deployment speedManual, slow, fragile60% faster, automated
    Production deploymentManual, real downtime riskZero downtime, automated
    HIPAA and GDPR readinessNot compliant, blocked from launchFully compliant, audit-ready
    Team working practicesData science, dev and ops all working separatelyOne shared pipeline, shared visibility

    How we measured it: deployment speed and downtime tracked from GitHub Actions run history before and after go-live.

    Business Impact

    Compliance stopped being the launch blocker. The startup went to market with an auditable, documented compliance posture instead of an open regulatory risk, and every future model update ships through the same controlled pipeline instead of a one-off manual push.

    What the client said

    Stonetusker helped us build pipelines that let us deploy AI models quickly and securely, which is critical in healthcare. Their guidance on HIPAA and GDPR compliance made us confident as we prepared for launch.

    Founder US-Based AI Cancer Detection Startup

    Building AI products in a regulated industry? Talk to Us About Secure Delivery See our Secure Delivery approach

    Why These Numbers Hold Up

    Every metric on this page is measured against a baseline captured before we start work, pulled from CI/CD logs, deployment history, or audit-trail timestamps the client already had. We agree on the baseline and the measurement method with the client in Week 1, then measure the same way at the end. That way, the before/after comparison is apples to apples, not a marketing estimate. Client names are withheld under NDA; every client featured here has agreed to act as a reference call for a serious prospect. We found that it takes 90 days to have a complete migration, automation & stibilize it and it reflects as RoI.

    Common Questions

    Why don’t you name your clients?

    Every case study on this page is published under a client NDA. We can’t use company names or logos without permission, but every client featured here has agreed to speak as a reference on request.

    What happens after I book a discovery call?

    It’s a 30-minute call with no pitch deck. We sign an NDA before we discuss your architecture, then talk through the problem you’re trying to solve and whether a 2 to 3 week paid pilot makes sense.

    What’s included in the 2 to 3 week paid pilot?

    Week 1 establishes a baseline and maps the priority bottleneck. Week 2 builds one production-relevant automation or workflow improvement. Week 3 delivers the measured before/after result, a handover, and a roadmap for scaling it.

    How do you measure results like “8x faster builds” or “50% more deployments”?

    We agree on a baseline metric and a measurement method with you before we start, typically pulled from CI/CD logs, deployment history, or audit-trail timestamps you already have. The same method is used to measure the result at the end, so the comparison is apples to apples.

    Do you only work with large, regulated companies?

    No. We work across three specializations: embedded and IoT hardware companies, regulated industries such as healthcare and fintech, and scaling SaaS companies. Company stage matters less than whether the engineering problem is expensive and time-sensitive.

    What happens after the pilot if we want to continue?

    If the pilot proves out, we scope a larger engagement around your priority roadmap. There’s no auto-renewing retainer; engagements stay milestone-billed.

    Got a similar problem?
    Let’s talk about it.

    Pick one expensive engineering bottleneck. In 2 to 3 weeks, we establish the baseline, automate or improve the workflow, and show you the measured change before you commit to anything larger.

    WeekFocusDeliverableSuccess signal
    Week 1DiagnoseBaseline metrics, architecture/workflow map, priority bottleneck identifiedAgreed baseline metric
    Week 2BuildOne production-relevant automation or workflow improvement, shippedWorking change in your target environment
    Week 3ProveBefore/after result, handover, roadmap for scaling itMeasured improvement + next-step plan
    Book a Free Discovery Call Get Your Free DevOps Health Score

    30-minute call  ·  No pitch deck  ·  We sign an NDA before we discuss your architecture