Skip to content
Avatar of Janne Kavander

Lead Software Engineer

to

I was the primary technical owner of Process Intelligence, ProcessMaker's process and task mining platform, after its acquisition. I was the only person left with deep knowledge of the product, having built it at the startup ProcessMaker acquired, so the platform's survival started with me: I rebuilt the engineering team, recreated the missing development and staging environments, stabilized deployments, and supported the relaunch within two months. From there I ran the platform end to end for over two years: roughly 70% hands-on engineering, the rest formal Engineering Manager work for up to four direct reports.

The platform

Process Intelligence combined process mining and task mining: it collected activity data from explicitly whitelisted applications, aggregated it every 24 hours, and showed organizations how work actually moved through their business processes, where the bottlenecks were, and what could be automated. It ran for large enterprises in banking, insurance, logistics, and manufacturing, with hundreds of data-collection agents reporting every five minutes and 100M+ raw MongoDB documents condensed into tens of millions of analytics records by the aggregation algorithms behind the product. I owned it as a complete system: backend, frontend, data collection, pipelines, authentication, infrastructure, CI/CD, monitoring, and integrations.

Reliability and architecture

  • Eliminated a daily platform-wide availability failure by redesigning authentication and request routing: long-running operations were blocking a shared Django worker pool, so a single user could exhaust capacity. After the redesign shipped, the daily scaling incidents stopped entirely.

  • Cut production APIs from 14 to 9 and repositories from 40+ to the low 20s, simplifying a microservice architecture built for a bigger team than we had. Azure Artifacts feeds went from 6 to 2, with shared data models, authentication, MongoDB testing utilities, linting and reusable libraries centralized along the way.

  • Removed an internally maintained fork of a testing framework the team had been carrying, along with the duplicated tooling that had grown up around it.

  • Resolved customer-facing data inaccuracies by refactoring critical MongoDB data models and aggregation pipelines, and removed incompatible duplicate model definitions.

Infrastructure and environments

  • Created the development, staging, and sandbox environments the platform never had, built on Bicep templates plus substantial manual configuration, with anonymized production data where realistic testing needed it. Deployments could finally be validated before production, defects investigated against real datasets, and changes previewed with customers and partners. The sandbox was later retired when its running cost outgrew its value.

  • Built out two new Azure production regions (East US and Australia Southeast) for enterprise data-residency requirements, taking the platform to three in total. Each one is a full production stack rather than a copy of a service: containerized app services and functions, databases and storage accounts, virtual networking and data routing, key vaults and access policies, gateways, monitoring and alerting, and its own deployment pipelines.

  • Kept ownership of the entire Azure estate even after onboarding CloudOps through handover workshops: containerized App Services and Functions, Container Apps, storage accounts, Key Vault, virtual networks, Application Gateways, IAM and access policies, MongoDB infrastructure, SQL Server, and the configuration holding it together.

  • Designed a cost redesign projected to cut annual Azure spend by roughly 60%, about €108,000 a year, implementation-ready when the role ended.

  • Contributed technical controls and evidence to a successful SOC 2 Type 2 certification: access controls, private networking, logging, secrets management, and dependency hygiene.

Delivery and developer experience

  • Caught data and pipeline failures before customers reported them, by rebuilding monitoring from a deprecated state into Grafana dashboards and Azure alerting over raw collection volumes, aggregation health, pipeline failures, and regional environments. Watching raw input against aggregated output surfaced problems in either layer.

  • Cut CI/CD pipeline times from ~20 to ~3 minutes, and failure rates from ~70% to under 2% in some repositories, through shared workflows, caching, smoke tests, and automated versioning. Smoke tests caught deployment defects before customers saw them more than once.

  • Took merge conflicts from constant to almost zero by establishing the engineering standards the team didn't have: a defined branching model, branch protection, mandatory reviews, required tests, conventions for how features are developed and promoted between environments, and automated versioning.

  • Supported a two-week release cadence with same-day hotfixes by defining release and QA processes end to end, from versioning and acceptance criteria to hotfix handling.

  • Cut engineer ramp-up from one month to 1 to 2 weeks with an onboarding system of architecture and service documentation, recorded walkthroughs, workshops and guided ownership handovers, used for engineers, QA, and CloudOps alike.

Product, frontend, and integrations

  • Embedded our analytics into a platform serving 500+ enterprise customers, designing and largely implementing the ProcessMaker 4 integration with secure tenant and session handling, replacing a costlier and more limited analytics approach.

  • Architected the ARIS Process Mining integration, feeding our process and task mining data into their platform to be visualised there. It shipped as a professional partnership, was presented at conferences, and put the product in front of ARIS's own sales organisation and enterprise customers.

  • Connected process discovery directly to automation with the process-to-workflow export: discovered processes could be converted into executable BPMN workflows in ProcessMaker.

  • Reached macOS customers the desktop task miner could not support at all, co-developing a production Chromium extension for privacy-first browser task mining. Publicly released, paid for by customers.

  • Gave customers visibility into their own data collection, owning the React dashboard's architecture and adding monitoring views for the health of their collection and analytics.

  • Rebranded the entire product after the acquisition with no customer disruption: application names, domains, UI assets, APIs, infrastructure, and deployment configuration across most services and repositories, on a tight deadline.

Team, stakeholders, and customers

  • Managed five engineers over the role, up to four at once, while remaining the platform's most active contributor: one-to-ones, performance reviews, compensation discussions, hiring, and career development. Also led QA, CloudOps, consultants, and adjacent-team engineers indirectly.

  • Ran or participated in 20 to 30 engineering interviews, both technical and personal.

  • Translated quarterly goals into two-week delivery cycles in Product Increment planning with 30 to 40 participants, onsite in the US and Bolivia or remote, renegotiating scope when priorities moved.

  • Investigated, fixed, validated and released customer escalations the same day, handling them weekly across data quality, deployments, and analytics requirements.

  • Pushed back where the business case didn't hold, including a proposed major expansion of the browser extension whose expected value didn't justify the investment.

  • Kept the team deliberately lean so more of its time went into building rather than coordinating, using Claude Code and Cursor daily. AI-assisted development is a large part of how one person stayed 70% hands-on while running everything above.

The through-line: I inherited a platform that had lost its team, its documentation and its stability, and I left it reliable, SOC 2 certified, running in three regions, and shipping every two weeks with a team of four.