Trace

Inspiration People change and people leave, that was the motivation point for Trace: leaving a trace of a person. AI today is built to answer almost anything, but far less attention has been paid to the idea that language models could also learn and preserve the way a specific person once speaks, writes, and thinks.

What it does

Trace is a trainable personal model. It lets users upload personal material, turn it into structured memory, generate responses in their style, compare different outputs, and correct the system when it gets something wrong. Over time, the goal is for Trace to feel less like a chatbot and more like a learned extension of the user.

How we built it

We built the backend and frontend in JAC The backend handles memory, persona, scoring, and preference learning, while the frontend gives users a place to upload material, train the model through side-by-side comparisons, explore memories, and interact with Trace through a conversational interface.

Challenges we ran into

The biggest challenge was defining the product clearly. We kept having to choose between building a productivity agent and building a memory-and-persona system, and those are very different directions. We also had to stay disciplined about what to make real first and what to leave for later, especially because low-latency video generation is difficult to run well in local environments and expensive to do online, which forced real tradeoffs in what we could build.

Accomplishments that we're proud of

We built the core learning loop where users can provide material, get persona-shaped outputs, give feedback, and gradually refine the model also . We are also proud that Trace has a clear point of view. It is not just another AI wrapper, but an attempt to make something more personal, persistent, and human.

What we learned

We learned that sounding like someone is only the surface. A meaningful personal model needs memory, correction, preference learning, and structure underneath it. We also learned that strong product focus matters more than adding features that look impressive but do not strengthen the core idea.

What's next for Trace

Next, we want to make Trace fully usable end to end, deepen the memory and reflection experience, and keep moving toward a stronger memory-first, persona-first system. Long term, we want it to feel like a true digital extension of a person, something that does not just respond, but remembers, reflects, and grows with you.

Built With

• jac
• python