Resequence

Learn from today. Improve tomorrow—with evidence, not generic rules.

Comment

Inspiration

Resequence began with a problem I experienced personally: I struggled to stop using social media and was easily distracted once I opened my phone. What started as a quick check could turn into hours of scrolling, leaving less time and energy for the things I actually wanted to accomplish.

Many people face the same problem. Social platforms are built around fast, repeated rewards that can create powerful dopamine-driven habits. Most productivity advice simply says, “Use your phone less,” but that does not explain why the habit happens, how it affects the rest of someone’s day, or what they can realistically do next.

I built Resequence to provide something more useful: personalized, evidence-informed guidance based on what a person actually did. Instead of judging screen time or applying the same rule to everyone, Resequence studies the timing, duration, and order of activities to help users recognize distracting patterns and replace them with practical actions that fit their lives.

What it does

When Resequence opens for the first time, the user enters their name, selects their main productivity challenge, and describes what they want to change. This gives the analysis a personal target.

The daily experience contains four steps:

  1. Priorities: The user chooses what matters today and rates each task's importance and difficulty. AI does not decide what counts as productive.

  2. Map your day: Quick Capture turns a natural sentence containing an activity, time, and duration into a clean timeline entry. Activities can be renamed, resized, connected to priorities, and swapped without changing unrelated activity times.

  3. Outcomes: The user reports how much of each priority they completed. Partial progress counts, producing a more honest measurement than a completed-or-incomplete checklist.

  4. Resequence: Before analysis, the user can inspect the evidence library and enable or disable individual sources. The AI may only use enabled evidence.

The final debrief explains what worked, identifies possible friction, and proposes one small experiment. It also provides three flexible suggestions for tomorrow. Each suggestion includes an event-based anchor, a success measure, a confidence level, and links to its supporting research.

Resequence does not invent tomorrow's commitments or generate a rigid schedule.

How I built it

I built Resequence with Next.js 16, React 19, TypeScript, and custom responsive CSS. It is deployed on Vercel.

I deliberately separated deterministic calculations from AI interpretation. TypeScript calculates weighted task progress, priority-linked minutes, context switches, awake time, and the day score. The model interprets those facts but cannot replace or contradict them.

The daily analysis uses the OpenAI Responses API with GPT-5.6 Terra and strict structured output. The server validates the returned structure and removes citation IDs that are not enabled by the user.

Quick Capture uses GPT-5.6 Luna to extract a clean activity name, start time, and duration from natural language.

The evidence library contains ten curated sources covering interruptions, task switching, notifications, screen-time friction, procrastination, breaks, exercise, sleep, implementation intentions, and chronotype. Each source records:

  • claims it can support;
  • claims it cannot support;
  • study limitations;
  • confidence and population information; and
  • small experiments the AI may recommend.

The MVP stores profiles and daily records in browser localStorage. This avoided unnecessary authentication and database complexity during a two-day hackathon. Deterministic fallbacks also keep Quick Capture and analysis functional if the AI service is unavailable.

Challenges I ran into

Keeping the user in control.
I did not want the AI to decide which activities were productive. The user defines their priorities, importance, difficulty, and completion, while the app calculates the results transparently.

Making timeline editing predictable.
Moving activities originally shifted every later activity. I redesigned the interaction as a direct two-activity swap so unrelated times remain unchanged.

Preventing generic AI advice.
A normal prompt could easily produce unsupported rules such as avoiding phones every morning. I created a structured evidence library with explicit supported claims, unsupported claims, and limitations.

Avoiding a thin AI wrapper.
The model does not calculate the score or freely invent conclusions. Deterministic code produces the facts, the evidence library constrains the available claims, and the server validates the structured response.

Choosing the right hackathon scope.
I considered adding Supabase and automatic scheduling, but neither was necessary for the core experiment. Browser-local persistence and flexible suggestions created a more focused and reliable MVP.

Accomplishments that I'm proud of

I am proud that Resequence is a complete end-to-end product rather than an isolated AI feature.

The MVP includes personalized onboarding, task prioritization, Quick Capture, an editable timeline, partial outcome tracking, deterministic scoring, ten inspectable research sources, structured AI analysis, evidence-linked suggestions, local persistence, dark mode, and a one-click demonstration.

I am especially proud that every recommendation remains explainable. Users can see which source supports it, understand that source's limitations, and disable it if they do not want the AI to use it.

What I learned

I learned that AI is most useful here as an interpreter, not as the authority deciding whether someone had a good day.

I also learned that personalization requires more than inserting a user's name into a prompt. The analysis must begin with the problem the user wants to solve and connect that goal to observable events and outcomes.

Finally, I learned that reducing scope can improve a product. Removing accounts, cloud storage, and automatic scheduling allowed me to focus on the interaction that makes Resequence different: turning a real day into a small, evidence-aware experiment.

What's next for Resequence

The next version could include:

  • helpful or not-helpful feedback for each suggestion;
  • multi-day pattern analysis;
  • opt-in accounts and cross-device synchronization;
  • user-approved calendar context;
  • additional research populations and productivity topics;
  • exports that users can share with a teacher, coach, or parent; and
  • personalized experiments that adapt based on previous results.

The long-term goal is not to produce a perfect schedule. It is to help people understand their own patterns and make one better decision at a time.

Built With

Share this project:

Updates

Submission history