Milestones

Inspiration

I received my first PhD application rejection email. In that instant, I didn't just lose an opportunity; I suddenly lost my footing.

I looked at my current productivity tools—my habit trackers, my streak counters. They were designed for the "Best Case" version of me, celebrating when I succeeded. But in the face of failure, they were silent. They offered no support when I felt like my effort had been wasted because the result wasn't what I wanted.

I realized that we have endless tools to optimize efficiency, but almost none to protect resilience. Milestones was born from this specific pain. It is designed to catch you during the "Collapse Moment"—that specific point where effort doesn't equal result, and you don't know if you should keep going.

What it does

Milestones is an AI-powered safety net that replaces the traditional "to-do list" with a "resilience engine". It functions through four key stages:

• Collapse Detection: The system monitors for behavioral signals (like silence for 7 days) or linguistic markers of despair (e.g., "I can't do this") to automatically trigger intervention.

• Glimmer Mode (The Sanctuary): When a collapse is detected, the UI transforms. It dims, plays calming white noise, and hides all pending tasks to stop the guilt cycle.

• Evidence Rebuild (The Constellation): Instead of a checklist of failures, the app visualizes a "Star Trail." The AI mines your history to prove you haven't given up, marking every tiny action—even just opening the app—as a shining star in a constellation of effort.

• The Non-Zero Action: To break paralysis, the AI proposes a ritual so small it is impossible to fail (e.g., "Open the document and close it immediately"). This proves to the user: "I am still here".

How we built it

We utilized a Resilience-First Stack designed to prioritize context and empathy over raw speed.

• Dual-Model AI Strategy:

• Gemini 1.5 Flash: Used for real-time interaction and quick action proposals.

• Gemini 1.5 Pro: The brain of the "Resilience Protocol." We inject the user's full 7-14 day task_history into the context window, allowing the AI to generate "Evidence-Based Hope" rather than generic advice.

• Frontend Visualization: We used Canvas 2D to render the "Constellation," turning data points into a visual safety net that connects every small attempt the user makes.

• Backend: Firebase/Firestore tracks the behavioral logs required for the Collapse Detector to function.

Challenges we ran into

• Suppressing "Toxic Positivity": The hardest part of Prompt Engineering was teaching the AI to not fix the problem immediately. We had to strictly forbid phrases like "Don't worry" or "Try again," which sound like mockery during a breakdown.

• Defining "Effort" vs. "Outcome": We had to build logic that distinguishes between what users control (the effort) and what they don't (the acceptance letter). We created an "Effort-Outcome Analyzer" to ensure the AI validates the former without promising the latter.

• The Liberation Question: It was risky to program the AI to ask, "Do you really want this, or do you just feel you should do this?" We had to carefully tune the tone to ensure it felt supportive rather than confrontational.

Accomplishments that we're proud of

• Visualizing Anti-Fragility: We successfully moved away from the "strike-through" list format. Seeing a user's struggle visualized as a "Constellation" of stars proved that they were still moving forward, even in the dark.

• Evidence-Based Hope: Our AI doesn't give empty compliments. Every word of comfort is backed by data from the user's history (e.g., "You showed up 6 times this week"), making the encouragement objective and undeniable.

• The "Ritual" Mechanic: We are proud of the "Non-Zero Action" feature. We proved that a 30-second action can be a valid milestone that helps rebuild self-trust.

What we learned

• Resilience > Efficiency: We learned that the most valuable feature isn't helping people work faster, but helping them not hate themselves when they fail.

• "At Least" is a Forbidden Phrase: We learned that using "at least" devalues effort. The AI must strictly state facts about effort without qualifying them.

• The Power of Reframing: Simply renaming a "task" to a "ritual" completely shifts the user's psychological state from pressure to permission.

What's next for Milestones

• Failure Rehearsal: Allowing users to pre-write a "comfort script" for their future selves, so the AI can deliver their own words back to them when they face rejection.

• Contrast Journaling: Automatically generating "Then vs. Now" perspectives to show users their long-term growth even when they feel stuck in the moment.

Built With

• aistudio
• english