Building AI systems that simulate human behavior is a technical challenge and an exploration of what drives us as people. As the head of engineering at Vurvey, my work centers on creating human behavior models that help us predict and understand complex market dynamics. These models function as living systems built from real human data. People create digital versions of themselves, and those models learn from their choices and stories.
⸻
⸻
At first I treated these models as tools to study how decisions unfold across many people. The work changed once I built a model of myself and started talking to it. Combining my model with those of colleagues and participants revealed something deeper. We were simulating individuals and watching people influence one another while adapting inside shifting environments. The aim of these People Models is simple and ambitious. Capture real stories, motivations, experiences, opinions, and the small nuances that generic systems miss. Acknowledge individuality rather than flatten it.
Working this way taught me about human behavior at scale, and it taught me about myself. Some lessons felt obvious in retrospect. Others surprised me enough to change how I see my own identity.
⸻
- Human behavior models challenge your self-perception
I have always thought of myself as careful and rational. I weigh options. I avoid unnecessary risk. Then my model predicted I would walk away from a safe investment and choose a speculative one. I felt defensive at first. That is not me. Except it is. Years ago I left a comfortable role at a large company for a startup. It did not feel reckless at the time. It felt necessary.
The pattern was there all along. I tend to choose purpose and growth over stability when the two are in tension. I would not have described myself that way before. The model did not change who I am. It sharpened the picture.
⸻
- Decisions are stories we tell about ourselves
The most revealing part of this work is the why behind each choice. These models are trained on decisions made by real people. When I asked my own model to explain a choice that looked irrational, it traced the decision back to a narrative I carry. I want to take meaningful risks for meaningful impact. That story can outweigh the neat logic of spreadsheets.
Often the model felt like a conversation where I already knew the answer but could not say it. The work helped me see how identity and self talk guide behavior, especially under pressure.
⸻
- Context and other people shape more than we admit
People rarely decide in isolation. We watch each other. We nudge each other. Our models simulate those interactions across segments, and the results mirror real life. Risk averse agents became bolder when surrounded by confident peers. Small shifts in one group set off larger changes elsewhere.
Seeing this in the system made me think about my own choices. Who I am with matters. My environment and my community amplify or dampen my tendencies. I notice the same effect at work, at home, and in my friendships.
⸻
- Feedback loops reveal the person and the system
Refining these models depends on feedback. The system predicts, reality responds, and we learn. Sometimes the model misses. Those misses are not only model errors. They are openings into the human side.
When my model failed to account for a pause before a big decision, it was because I had not acknowledged my own hesitation. The loop surfaced a contradiction I had glossed over. At a broader level, repeated loops exposed patterns I would have missed, such as how minor changes in group mood ripple into large outcomes.
If you enjoy thinking about loops and self reference, you will find the same theme in books like GEB and I Am a Strange Loop. The idea keeps showing up in different domains.
⸻
- Change is the rule
People evolve. Priorities shift. New experiences rewrite old habits. Early versions of our models struggled with that reality. They leaned too hard on yesterday. We needed systems that learn continuously so they can grow with the people they represent.
This lesson lands close to home. I notice how often I run on outdated assumptions about myself. It is easier to act from an old script than to face who I am now. The work keeps nudging me to update the script.
⸻
- Imperfection is where the truth hides
I once thought the goal was perfect prediction. The closer we got, the more I changed my mind. The most interesting insights often live in the inconsistencies. Human decisions can look messy. That mess carries information.
Treat contradictions as signals. They reveal values in conflict. They reveal emotion in the loop. They reveal what we are willing to trade when the moment finally arrives. Aim for understanding and respect rather than ironing every wrinkle flat.
⸻
Final thoughts
Building human behavior models has been a technical project and a personal one. These systems can forecast choices and surface the narratives, tensions, and relationships that shape a life.
If there is a single lesson I will take into the next year, it is this. Choose understanding over perfection. Build systems that learn from real people in real conditions. Let them change with the world. And when they show you something uncomfortable, pause and listen. That is where growth begins.