Kickly: תכנון וביצוע MVP מבוסס AI

MicrocosmWorks built the AI engine to support OpenAI GPT, Anthropic Claude, and Google Gemini as interchangeable generation backends with automatic fallback resilience. Users can select their preferred model, or the system defaults to the optimal model for the task type. Each idea generates 9 plan variations across 3 tiers (MVP, Growth, Scale) and 3 versions per tier, allowing founders to compare AI-generated approaches before committing to a development path.

MicrocosmWorks implemented real-time collaboration using Yjs CRDT (Conflict-free Replicated Data Type) and HocusPocus WebSocket server. Multiple team members can edit BlockNote rich text documents, Excalidraw whiteboards, and Mermaid diagrams simultaneously with live cursor presence and conflict-free merging. Changes propagate in real time without data loss, even when two users edit the same section simultaneously.

MicrocosmWorks built a cross-platform Electron desktop application that tracks work time with automatic screen capture at configurable intervals and idle detection that pauses tracking when the user steps away. Captured work logs sync to the Kickly web platform backend in real time, feeding into task-level timesheets, weekly reports, and team analytics dashboards. The desktop app provides accountability and billable-hour tracking for distributed startup teams.

MicrocosmWorks architected six NestJS microservices with clear domain separation: Auth (passwordless OTP with dual JWT), Project (idea management, plan generation, task hierarchy), Work Tracking (screen recordings, time logs, timesheets), Document (BlockNote editor, Mermaid diagrams, PDF viewer), Collaboration (Yjs/HocusPocus real-time sync), and Notification (email via AWS SES, in-app alerts). This separation enables independent scaling and deployment of each service.

MicrocosmWorks integrated Razorpay for subscription billing with full lifecycle management including checkout, plan upgrades, cancellation, and invoice generation. Plan tiers control the number of active projects, AI generation credits per month, team member seats, and storage limits. Credit consumption is tracked per AI generation request, with the system preventing over-usage and prompting upgrades when limits are approached.