Rafał – MongoDB, Node.js, React

A boutique software house delivering long-term EdTech and SaaS engagements for US and UAE clients, with dedicated engineering teams that own a client's product from architecture and delivery to production support.

• Co-founded the company and served as senior engineer on the flagship K-12 EdTech engagement (React, TypeScript, Node.js, MongoDB) while running the business side.
• Owned technical delivery standards: code review, test coverage (Jest, Playwright, Cucumber), CI/CD on CircleCI and Docker-based environments across client projects.
• Recruited, onboarded and mentored engineers for client teams; ran sprint planning and estimation with client product managers.
• Managed contracts, budgets and settlements with international clients and a distributed developer team.

A data platform for around 20 US school districts and 500 schools that gives teachers, counsellors and district leaders one place to track student progress and career-and-college readiness, with early-warning indicators, an intervention bank, progress monitoring and custom reporting.

• Owned architecture and delivery across the full stack: React/TypeScript front end, Node.js and Meteor back end, MongoDB as the single source of truth; reviewed and merged the majority of the team's pull requests.
• Modernised a legacy Meteor codebase to React and REST without a rewrite while the product stayed in production; led the Meteor 2 → 3 (async API) migration in 2025 with a controlled rollback around an upstream bug, then Meteor 3.4 with Rspack; drove the module-by-module move to TypeScript (2020–23).
• Cut initial load time from 40–50 seconds to about 5 by replacing heavy Meteor pub/sub with targeted REST endpoints and client-side caching, while keeping real-time editing for the screens that needed it.
• Replaced a forked in-house copy of Highcharts with the upstream library (~22k lines removed) and built a generic chart generator (factory/adapter patterns) covering 20–30 charts.
• Designed and built Azure Data Factory pipelines and a strictly validated REST import layer so no malformed student or staff data reached production; owned the 5,000+ line OpenAPI spec for the unified API over a shared SQL district system.
• Migrated the entire end-to-end suite from Cucumber to Playwright without stopping the team, via a Playwright-compatible driver layer that kept existing page objects and step definitions working.
• Kept the platform stable and shipping with a three-engineer team for the last five years (down from ~20); ran sprint planning, backlog grooming, code review and mentoring.

Nightly integration of student and staff data from a large SQL system shared with other school applications into a K-12 EdTech platform's MongoDB, serving around 20 US school districts and 500 schools. It replaced an earlier setup where unvalidated data reached production and caused reports to crash.

• Designed and built Azure Data Factory pipelines running nightly, with error handling, Azure logging and Teams alerts on failures.
• Built a strictly validated REST import layer: malformed records (missing IDs, wrong lengths, broken references) were rejected at the boundary and reported to the client's data team with the exact student and field to fix.
• Held the validation line against pressure to loosen it, on the evidence that bad data had previously broken the UI; the number of "app bug" reports that were really data issues dropped accordingly.
• Kept MongoDB as the platform's single source of truth while the shared SQL system remained the district's system of record, with a clear contract between the two.

A single unified REST API — required by the client's DevOps as one API with one OpenAPI specification — exposing the data a K-12 EdTech platform needed from a large SQL district database shared with other school applications.

• Mapped the required SQL structures and designed the endpoint layout for a spec that grew past 5,000 lines of YAML.
• Generated and maintained the spec with Cursor working against the schema, then verified it with spec validation and contract tests against the real database; caught invented fields and wrong nullable types before they shipped.
• Established the working rule the team kept afterwards: AI-generated config isn't done until a machine has checked it against reality.

A performance programme inside a K-12 EdTech platform used by around 20 US school districts, started after admins reported 40–50-second initial loads. The root cause was Meteor's pub/sub model pushing whole collections into the browser.

• Diagnosed the bottleneck and designed the migration path: moved state to Redux first, then replaced heavy subscriptions with targeted async REST endpoints returning only the data the current view needed, with selected slices cached in React Context.
• Reduced initial application load from 40–50 seconds to about 5 seconds; heavy reports and comparison views became near-instant.
• Preserved real-time collaboration where it mattered with a handful of narrowly scoped, paginated subscriptions and custom server-side observers, so multiple teachers could edit the same records concurrently.
• Measured every step with browser network timing and time-to-interactive on real district accounts.