Project
Tasty CRM & Workstream
An enterprise CRM used by 1,000+ people, built with a worldwide engineering team I lead. My work spans product ownership, platform architecture, Workstream task management, agent access through MCP, financial operations, and a largely self-hosted production stack.
Product ownership and a worldwide team
Tasty CRM is the enterprise platform I help build and operate with a worldwide team of 15+ engineers. More than 1,000 people use it. It connects creator operations with the financial and collaborative work behind the business, so a change in one module often has consequences elsewhere.
I designed the platform architecture. I still write its code. Leading a team across countries and time zones also means running standups and working with project managers and QA to carry a requirement through implementation, review, and release. That means deciding what needs to ship, reviewing the implementation, and following up when production exposes an edge case the original requirement missed.
Workstream inside the CRM
Workstream is our in-house alternative to Linear. I build and extend it as part of the CRM, where task ownership can use the same employees and team permissions as the rest of the company.
The work goes beyond a task list. Projects and task dependencies support delivery planning. Bulk actions handle changes across many tasks, while subscriptions and comment discussions keep the people involved informed. I have worked on project-manager access, shared assignments, and the details that make long task conversations usable, including mentions and comment navigation.
Keeping Workstream in the CRM also puts the engineering workflow beside the operational work it supports. A task can carry its discussion and history through implementation and QA without requiring the team to recreate that context in another product.
Agents use the same working system
I built the Workstream MCP server in April 2026. It gives compatible agents authenticated tools to read and update tasks, along with access to projects, comments, and activity. Per-user permissions and audit logging remain part of those operations.
Since then I have worked on tenant isolation, retrieval overhead, and agent setup. The server is now being extended toward a broader CRM MCP. The first Finance Costs integration adds read-only access through the Costs query layer, with self-service key management and agent setup in the employee profile. Wider CRM coverage is still being developed.
Performance and concurrent work
I worked through Workstream's frontend, API, and MCP server to reduce the repeated queries and long-held transactions behind everyday task interactions. Grouped task views had fetched pages sequentially. That path now uses a single list query. Reordering now batches writes, and repeated permission checks are cached within each procedure call so unrelated mutations cannot share stale permission state.
I also removed unnecessary refetches from closed dialogs and kept external agent calls from holding a database connection while they wait. Each interaction asks less of the server.
The platform has real-time document collaboration through Yjs and Hocuspocus, with document state stored in PostgreSQL. Supporting simultaneous work also requires attention to query volume and background jobs.
Financial operations and production ownership
I built the finance module end to end, including multi-method payment processing and recurring costs, invoice generation, P&L reporting, and two-factor authentication around sensitive operations. Those workflows sit alongside creator management and internal collaboration in the same TypeScript monorepo.
The production stack is largely self-hosted, with Docker and Coolify, PostgreSQL, and separate services for collaboration and MCP. Prometheus, Grafana, and Loki provide monitoring and alerting. GitHub Actions handles delivery, including migration safeguards and checks before a pull request reaches release.
I also contribute to InfraAgent, our public infrastructure monitoring product, and help shape SEO direction and discoverability across the company's public products. That work connects the engineering decisions to how someone first finds a product and keeps using it.