LifeStrat

• 

  This could be another you.

Inspiration

As college students, we're constantly told to "figure out our future," to "stay on top of our goals," and to make the "right" decisions, but often without any clear guidance on how to actually do that in a way that feels personal, motivating, or even manageable. Even subscribing to career newsletters and other emailing lists can quickly get overwhelming, especially when it feels like so much is at stake. So, we decided to create an AI assistant that can cut through that noise.

LifeStrat was inspired by a simple idea:
what if your life didn’t feel like chaos — but like something you could see, understand, and grow?

What it does

LifeStrat is an AI-powered career companion — but at its core, it feels more like a game. At the center is a character that represents you. From an outside perspective, you are not just managing tasks — you are watching a life unfold.

• When you take action, the character grows.
• When you neglect areas, parts of their world go quiet.
• When you stay consistent, momentum builds.

Your schedule, goals, and opportunities are no longer abstract — they directly shape the character’s state. The system pulls real opportunities from your inbox, ranks them based on your goals, and helps you plan your time intelligently. But more importantly, it makes your decisions visible.

There are no failure states — only different seasons. Over time, the character evolves to reflect who you are becoming.

How we built it

We built LifeStrat as a full-stack web app using: • Next.js (App Router), TypeScript, Tailwind CSS • Zustand for global state (with persistence) • Supabase (PostgreSQL + Auth) for secure, per-user data • Gmail API to import real user emails • Zed as a code editor to build and iterate our project

We use the K2-Think-v2 model in three key places: 1 Email classification → turning raw inbox data into structured opportunities 2 Opportunity ranking → combining heuristics with AI semantic reasoning 3 AI scheduling → generating optimized daily/weekly plans based on constraints

Opportunities are scored using a layered system (urgency, effort, preference fit, etc.), then refined with AI to capture deeper career alignment beyond keyword matching.

Challenges we ran into

Our hardest challenge was building the AI planning and scheduling system. We needed the AI to: • avoid conflicts with fixed events, • balance workload realistically, • prioritize high-value opportunities, • and split tasks across multiple time slots when needed.

The difficulty wasn’t the model itself — it was teaching it how to think.
Early outputs looked reasonable but broke key constraints.

We solved this by designing structured prompts that clearly defined rules, tradeoffs, and priorities. We also struggled with personalized ranking. Simple keyword matching failed to capture real career alignment, so we built a hybrid system combining heuristics with AI reasoning.

Accomplishments that we're proud of

We’re proud that LifeStrat works on real user data, not just a demo. • We built a full pipeline from Gmail → AI classification → ranked opportunities → scheduled plans • We created a system that turns messy inbox data into clear, actionable decisions • We designed a character-driven experience that makes productivity feel engaging and personal

Most importantly, we built something that doesn’t just organize tasks —
it helps users see their growth over time.

What we learned

We learned that building an AI product is not just about the model — it’s about everything around it.

This was our first time working with an LLM API, and one of the biggest surprises was how much prompt design matters. Small changes in wording could completely change the quality of the output. Clear, structured instructions consistently outperformed longer, more complex prompts.

We also learned that debugging AI systems requires thinking in scenarios, not just code. We tested across different user accounts, built multiple profiles, and simulated edge cases to understand how the system behaves under real-world conditions. On the engineering side, we gained a much deeper appreciation for infrastructure challenges. Integrating the Gmail API forced us to understand OAuth flows and work through token limitations, while ensuring proper per-user data isolation required careful state and database design.

From a product perspective, we learned the importance of focus and intentional design. We had many ideas that were exciting to build, but we had to prioritize what actually mattered most to the user.

Most importantly, we realized that AI becomes truly powerful when it doesn’t just automate tasks —
but helps users see themselves more clearly and take ownership of their growth.

What's next for LifeStrat

We want LifeStrat to evolve from reactive planning → proactive, personalized coaching. • Smarter opportunity ranking, improving how we assign importance and relevance using better prompting and deeper personalization • More intelligent scheduling, incorporating real-life factors like commute time, attention span, energy levels, and user habits • A more conversational AI, moving from one-time plans to ongoing dialogue, where users can share goals and the AI adapts over time — even recognizing burnout and adjusting intensity • A more alive character system, making the character more present, expressive, and integrated across the entire experience. New ways to calculate character growth and change • Improved UI/UX, creating a more responsive, readable, and engaging interface • Open collaboration, exploring open-sourcing the project so others can contribute ideas and improvements

Long term, we want LifeStrat to feel less like a tool —
and more like a system that grows with you.

Built With

• css
• gmail
• google-gmail-oauth
• mbzuai-ifm
• next.js
• react
• supabase
• typescript
• vercel
• zustand