Building collaborative, AI-assisted
design-system workflows.
Connecting the design, the contribution process, and the documentation—across Figma and the web.
Design. Contribute. Document.
A component can look complete while its rationale, use cases, properties, and documentation remain unresolved.
Compfeed makes that work visible through contribution phases, tasks, and supporting notes. The Figma widget puts the process beside the design; the web application gives the team a place to develop and document it together.
I joined an existing product as a Backend Engineer, moving into a Full-stack Engineer role in June 2025. I built Playground and Manifest features, AI documentation, the custom Figma MCP integration, and component dashboard functionality.
I migrated Express.js to NestJS and set up Docker-based deployments. Other teammates initially owned the widget; I later took over its maintenance.
From a linked design to a document the team can refine.
Product detail · Select to enlarge.
A frame, section, or page becomes the design reference for a contribution. The checklist stays beside the work.
My contributionI later took over maintenance of the existing widget, including its data management and associated flows.
Design previews, phase labels, progress, and filters connect individual components to the wider contribution process.
My contributionI built dashboard functionality around linked design information.
The design preview, property inspection, phases, tasks, and notes share one workspace.
My contributionI built Playground features and coordinated collaborative content with dynamic workflow updates.
Contribution context and linked Figma information support a first draft that people review and refine.
My contributionI built Manifest and AI documentation functionality. The integration reached customer beta.
Collaborative edits and Firebase-driven reordering could overlap, inserting duplicate task and user-content nodes.
I protected the relevant editor-update operations with a mutex. Each operation acquired the lock, applied its changes, and released it before the next entered that path.
An incoming operation may wait while the current update completes. Coordinating these protected paths addressed the observed duplication race.
I tried tracking task, phase, and content IDs, adding delays or debouncing, and improving provider lifecycle and synchronization checks. They reduced symptoms without reliably coordinating all asynchronous editor-update paths. Explicitly protecting the boundary was the useful shift.
A queue organizes work. It does not tell a job whether it is eligible to run.
I persisted per-user cooldown timestamps after Figma returned HTTP 429. Background jobs checked this state before sending another request, deferring affected work while other users continued.
The affected user’s stored context can lag during cooldown. Background jobs avoid repeated requests during a known restriction while other users continue.
A Figma response is not automatically useful model context.
I built the custom integration across the NestJS MCP server, Figma retrieval, structured extraction, AI connection, and dashboard. The challenge was preserving useful relationships while bounding large, nested responses.
Scope retrieval to the linked frame, section, or page. Validate the reference and handle missing or unavailable nodes.
Validate tool arguments. Combine stored context with background, on-demand, and manual refresh—without webhooks. Balance freshness against API traffic.
Select relevant fields, traverse useful child nodes, and limit depth. Retrieve deeper details separately when the task needs them.
Split large outputs into manageable parts and fetch additional details on demand. Preserve relationships when assembling context.
Combine design information and contribution notes for AI-assisted drafting, followed by collaborative review and refinement.
I migrated the existing Express.js backend to NestJS alongside ongoing feature development. The value was a stronger foundation for the work that followed.
An existing product.
Features still moving forward.
A backend I could extend as my responsibilities grew.
A more useful development foundation for extending the product alongside the migration.
Made it easier for me to manage widget data as I later took over maintenance.
Support for changing UX flows as my work expanded across the backend and frontend.
Docker-based deployments to Heroku and Google Cloud Platform, with GitHub Actions among the project’s delivery tools.
The most demanding problems appeared where systems met.
Dynamic data entering a collaborative editor. Queued work meeting an external API limit. Complex design structures becoming model context.
Shipped Playground, Manifest, dashboard functionality, AI documentation, and the custom Figma MCP integration; later maintained the existing widget.
Addressed the observed collaborative duplication race by coordinating the protected update paths.
Added persisted cooldowns so affected background jobs could wait while other users continued.
Automated parts of collecting design context and preparing documentation drafts for review.
Improved the development foundation for widget data management and UX flow changes through the NestJS migration.
Delivered the connected AI workflow to customer beta, bringing Figma context into the contribution and documentation process.
Make coordination explicit.
Model restrictions as state.
Evaluate the answer, not just the retrieval.
Based on my engineering notes, clarifications, and the three supplied product screenshots. Public context: Compfeed, the contribution process, Figma Community widget, and LinkedIn.
Public pages were reviewed on 4 October 2026, after my involvement ended. They do not establish that every current feature existed during my tenure or was built by me. Screenshots are not dated release evidence. No adoption, revenue, or performance metrics are claimed.
Release and measurement scope. The AI integration reached customer beta; some answers were incorrect or irrelevant and required human review. Time savings, delivery-speed improvements, and runtime performance were not measured. The NestJS migration’s value is described from my development experience.
Engineering boundaries. The mutex coordinated protected update paths; a local mutex alone does not guarantee global distributed consistency. Cooldown handling is confirmed for background synchronization, not every interactive MCP request. The listed technologies span my work and did not all form one request path.
Visual evidence. Product captures are cropped to remove browser chrome and focus on relevant features. They are not dated release evidence or demonstrations of successful AI output. Manifest is a labeled schematic, not a screenshot or generated document. My involvement ended on 20 January 2026.