How srinivas917/taskingo works
Model the intended product architecture (goal/task tracking + AI assistance) clearly.
2) Leave implementation choices as placeholders (e.g., “Backend API (TBD: Node/FastAPI/etc.)”) until code is available.
Backend gathers context
• active goal
• open tasks
• progress history
• preferences/constraints
AI output is parsed into structured tasks
• Tasks
• reorder/prioritize
• propose deadlines
Reminder / check-in flow (worker)
• Scheduler triggers:
• daily summary
• overdue reminder
• “you haven’t progressed” nudge
• Worker fetches user schedule/preferences → sends notification
Architectural patterns to highlight (even before code exists)
Given the product description, call these out visually:
• Clean separation between Domain (goals/tasks) and AI integration
• AI should be an “advisor,” not the source of truth for task state.
• Event-like progress tracking
• Task state changes can be stored as events (optional), enabling analytics.
• Asynchronous background processing
• reminders, summaries, heavy AI tasks, embeddings, analytics.
• Auditability
• store AI prompts/responses (with redaction) for debugging and safety.
Technologies to include (how to label without guessing)
Since there’s no code to confirm stack, label with intent + placeholder:
• “Client App (Web/Mobile — TBD)”
• “Backend API (REST/GraphQL — TBD)”
• “Database (SQL/NoSQL — TBD)”
• “LLM Provider (OpenAI/Anthropic/Local — TBD)”
• “Notifications (Email/SMS/Push — TBD)”
What you need next to make it truly “most accurate”
Right now, accuracy is limited by missing code. To finalize:
• Actual folders/files once they exist (e.g., frontend/, backend/, services/ai/, db/, infra/).
• Confirm:
• single service vs microservices
• storage type
• whether AI is synchronous user-triggered vs scheduled
• whether there’s a queue (SQS/RabbitMQ/Kafka)
• auth approach
• Goal/task domain as the core
• AI as an integration with orchestration, parsing, and audit trail
• Background jobs for reminders/summaries
• Strong, explicit data flows (CRUD + AI suggestion lifecycle)
• Placeholders for stack choices until code appears
This gives maximum detail without guessing implementation details that aren’t in the repository yet.