Trace

Inspiration

We students and young professionals need to make important life decisions such as choosing a university, accepting a job offer, changing majors, moving abroad or doing a startup. These decisions are often made under pressure, or with incomplete information or confusion. There are existing AI assistants who generate advice within some time, but they answer the question without understanding the person. As a result, users receive different answers every time they ask and the AI never learns how they think. We felt that real issue with the current systems is not lack of information, but it is lack of structured thinking. People don't need AI to decide for them, but need AI to help them think better. That's how we came up with Trace. Instead of making another chatbot, we imagined an AI that grows alongside the user and understands it better. Rather than replacing human judgment, it helps people think more clearly, recognize hidden assumptions, explore realistic possibilities and learn from every important decision they make. We wanted it to become the second brain of the person.

What it does

Trace focuses on structured reasoning rather than giving direct advice or recommendations. The user begins by entering a decision into the system. This creates a new Trace, which acts as a guided thinking space for that specific choice.

The process then follows a clear 5-phase flow called the Trail.

First, in the Frame phase, Trace understands the decision context by asking a small number of adaptive, mostly multiple-choice questions. These questions help clarify goals, constraints, priorities and concerns. If the system already knows relevant information from the user’s past decisions, stored in the Imprint, it reduces or skips repeated questions. The result is a short, editable summary of the situation.

Next, in the Markers phase, Trace identifies the hidden assumptions behind the user’s thinking. These are beliefs that may not have been tested, such as assumptions about career satisfaction, financial stability or personal preferences. Each Marker is shown with its evidence level, confidence and importance. If the same assumption has appeared in previous decisions, it is highlighted as an Echo, showing recurring patterns in the user’s thinking.

In the Trail phase, Trace explores possible outcomes of each option. It shows how each path could unfold using probably, highlighting opportunities, risks and tradeoffs. It also compares options against the user’s priorities to show where each path aligns or conflicts with what matters most to them.

After that, in the Fieldwork phase, Trace suggests real-world validation experiments called Field Tests. These are small, practical actions designed to test the most uncertain and important assumptions. Users complete these in real life and then log their Findings, reflecting on what they learned and whether their assumptions were correct or incorrect.

Finally, in the Footing phase, Trace evaluates how well-supported the decision is based on completed Field Tests and remaining uncertainty. This produces a clarity score called Footing, which is not a recommendation but a reflection of how strong the evidence is for the decision at that moment. Once this phase is completed, the learnings from the entire Trace are folded into the user’s persistent memory system called the Imprint.

The Imprint is a long-term, evolving profile of how the user makes decisions. It stores values, risk preferences, decision habits and recurring patterns of the user. Over time, it helps future Traces become more personalized by reducing unnecessary questions and identifying repeated biases or assumptions.

Across all steps, Trace never tells the user what to choose. It only helps them understand their options, test their assumptions and make more informed decisions. The final decision always remains with the user.

How we built it

We built Trace as a modern web application using Next.js and TypeScript, with Tailwind CSS and Framer Motion creating a clean, responsive, and interactive user interface. State management is handled through Zustand, allowing every Trace and the user's personal Imprint to persist across sessions. The reasoning engine is powered by a Large Language Model through the Anthropic API. Rather than generating generic advice, carefully designed prompts guide the AI through a structured reasoning process that identifies tradeoffs, explores multiple scenarios, generates personalized Field Tests, and summarizes user reflections. For offline demonstrations and environments without an API key, the application includes a realistic demo mode that simulates the complete reasoning pipeline. The overall architecture follows a clear flow of user input, AI reasoning, structured outputs, reflection, and continuous personalization, making Trace feel less like a chatbot and more like an evolving thinking companion.

Challenges we ran into

One of our biggest challenges was designing an AI that supports human thinking without making decisions on behalf of the user. It was tempting to generate direct recommendations, but doing so would encourage over-reliance on AI and reduce thinking from the user's side. Another challenge was balancing simplicity with depth. We wanted the experience to remain approachable while still helping users explore complex tradeoffs in a meaningful way. We also worked to ensure that uncertainty was communicated honestly rather than presenting AI-generated outputs as facts. Finally, designing a system that learns from previous decisions while remaining transparent and editable required careful consideration of both user experience and responsible AI principles.

Accomplishments that we're proud of

Our proudest achievement is transforming a traditional AI assistant into a genuine second brain that evolves with its user. Unlike conventional chatbots that forget previous conversations, Trace continuously builds an Imprint that captures personal values and decision habits over time. We are also proud of the personalized Field Tests, which encourage users to gather real-world evidence before making major life decisions instead of relying entirely on AI-generated responses. Combined with adaptive questioning, scenario simulations, and transparent uncertainty indicators, these features create a system that supports thoughtful decision-making rather than replacing it. Most importantly, Trace keeps the user in control at every stage. The AI provides structure, insight, and reflection, but the final decision always belongs to the individual.

What we learned

While we spent our time building Trace, we understood that AI is most useful when it helps people think clearly and understanding the user instead of simply giving answers. We learned that asking the right questions and guiding users step by step leads to better decisions than just generating long responses. We also learned that responsible AI is very important. Our AI explains its reasoning and shows uncertainty. Also, It never makes the final decision for the user. The user is always in control. Then, we learned how good design can make difficult decisions feel much easier. By breaking a big problem into small, simple steps, we help users feel less overwhelmed and more confident in making their own decisions.

What's next for Trace

Our long term goal is to make Trace a lifelong decision companion that grows along with its users. In the future, we plan to add cloud syncing so users can access their account from any device, support collaborative decisions with friends or family, connect with calendars for planning Field Tests, and provide deeper insights into decision-making habits. We also want Trace to become more personalized over time by learning each user's values, goals and thinking style. No matter how much it improves, the final decision will always stay with the user.

Built With

• anthropicapi
• framermotion
• next.js
• openai
• reacthook
• tailwindcss
• typescript
• vercel
• zustand