AI Systems Engineer

What Balance is

Balance is a life design app. It helps people build lives they enjoy, find fulfilling, and find rhythm in. It does this one small, well-chosen moment at a time.

It learns from what people do, write, plan, and finish. It uses that to suggest what's worth doing next: a reflection, an outing, a habit, an event nearby, a better shape for the week. After two years of building, it's in limited testing and about to open to paying subscribers. Real data is coming in, and the open problems are no longer hypothetical.

The AI isn't a feature bolted onto Balance. It is the product. If the suggestions feel generic, nothing else matters. If they feel like they came from someone who knows you and has read the research, people come back. Your job is to make that second outcome happen reliably.

The role

You'd own the intelligence layer of Balance from end to end. That covers what we learn about a person, how it's stored, what context each prompt gets, which model runs it, and whether the result actually landed.

In plain terms: the right prompt runs with the right context on the right model, at a cost we can afford. And we have real ways of knowing the result was good: signals where they exist, judgment where they don't.

You'd work directly with the founder. You'd have real authority over architecture. This is an early-team role with real ownership, not an internship with a nicer title.

Problems you'd own

These are real, open, and in the codebase today. Pick where to start and make the case for it.

• Knowing when a suggestion landed. Build the evaluation layer: traces, graded example sets, model-as-judge checks, and the signals that tell us a suggestion resonated. Much of the rest of this list depends on it.
• Portraits. Turn scattered signals (journal entries, completed moments, plans, skips) into a living model of a person. The model should be specific enough to generate from and honest about what it doesn't know.
• Memory and context. Decide what each call should see, what to summarize, what to forget, and how to keep long-running context from drifting or bloating.
• Evidence-backed guidance. Build a corpus of research on wellbeing, habits, relationships, and meaning. Wire it in so the AI blends general wisdom with what it knows about this one person, and can say why.
• Background agents. Design the pipelines that read new data, update inferences, and prepare recommendations ahead of time. They should run reliably, be observable, and be cheap enough to run for every user.
• Routing and cost. Match each task to the right model and technique: a cached prompt, a small model, a big model, or plain code. Know when a deterministic rule beats a model call.
• Reading intent. Test whether newer approaches (structured intent models, causal reasoning about what moves someone, steering) belong in Balance, and prototype the ones that do.

Stack today: TypeScript, Postgres via Prisma, background job queues on Redis, Heroku, and several frontier LLM providers.

Your first month

We're not looking for patches. Most of the pieces work. What they need is someone who understands the whole system well enough to reshape it with care.

• Week one: learn how it really works. Trace what Balance knows about a person, every prompt it runs, and where each suggestion comes from. Find what's solid and what only works by luck.
• Week two: form a point of view. Write up and map out how you'd shape the system as a whole: what to keep and refine, what to rebuild, and why.
• Weeks three and four: start building it. Land the first pieces in production, finished well enough that people feel the difference.

Who you are

• You've built something real with LLMs, past the demo stage, and you have opinions about what broke.
• You have taste, and you respect evidence. Some of what makes a suggestion good can be measured, and some of it is closer to art. You can tell which is which.
• You can't stand AI slop. You notice generic, hedged, over-eager output and you treat it as a bug.
• You set your own direction. Give you a goal and context and you come back with a plan, a prototype, and the tradeoffs. You know when to lean on a proven pattern and when to invent one.
• You care about people more than about models. The test is whether someone's week actually got better.
• You write clearly. Most of this job is deciding what a model should know and saying it precisely. You can also draw how a system works, so others can see it and weigh a change before it's built.

Above all, you're reliable without being chased. You show up consistently, keep us posted without being asked, and say so early when you're stuck or behind. Going quiet isn't an option on a team this small.

Bonus if you've gone deep on any of these: agent harnesses, retrieval, evals, fine-tuning or steering, causal inference, behavioral science, positive psychology, learning science, experience design, or personal assistant and planning tools. A taste for design that feels warm and human helps. So does business sense, because every call costs money.

Why join

Your work shapes how the whole product thinks. You'd own a core system early, set its standards, and have a path to leading AI at Balance as the team grows.

We value clarity, curiosity, and play. We take the work seriously, not ourselves.

Equity and time

We can't pay a salary yet. We're raising, and this role gets the first salary when we do. Until then, it works like this:

1. 1A trial period first. You pick one problem from the list above and own it. We both find out whether this is a fit.
2. 2Equity, vesting from your first day. A written grant, sized to the role and to what you bring, vesting over four years with a 6-month cliff. Your trial counts toward it.
3. 3Salary when we raise. Market-informed pay for this role is part of the plan for the round, not an afterthought.

Time: 20–40 hours a week to start, full-time once there's salary. Location: remote (U.S.-friendly hours), with occasional in-person collaboration. Openness to relocating later is a plus.

We know equity-only is a real ask. It suits someone who has runway, is compelled by these problems, and cares about building experiences that are genuinely good for people. We have high aims for this company and are building something with broad appeal. The ceiling is high but this role doesn't suit someone who needs income this month.