TidyMind — AI Task Prioritization
Turning a messy brain-dump of tasks into a reliably ordered plan with LLMs.
Engineer · Side Project
The problem
People usually know what their tasks are but stall on what to do first. Naively asking an LLM to "prioritize my todos" produces plausible-sounding but inconsistent, unparseable answers you can't build a real interface on.
Goal & constraints
Build a task app where an LLM reliably prioritizes tasks and returns structured output the frontend can trust every single time.
Key decisions
Engineer prompts for structured, schema-shaped output instead of free-form text.
The UI needed dependable fields — priority, ordering, rationale — so model output had to be parseable, not prose. Prompt design was the actual product surface.
Add a response-parsing layer that validates and recovers from imperfect output.
LLMs drift. A parsing and validation step kept a single bad response from breaking the whole task list.
How I built it
Prioritization engine
Built the LLM-backed flow that ranks a user's tasks and explains the ordering.
Reliable outputs
Designed prompt engineering and response parsing to produce reliable structured outputs the app could render directly.
Outcome
A working AI task-prioritization platform where the model's output is structured and dependable enough to drive the interface.
What I took away
- Reliability with LLMs is mostly an engineering problem around the model — prompt shape and output validation — not the model itself.