Software Engineer, AI Engineer (Applied​/Software)

LLMs are trained to follow the human they're talking to. Ours has to lead: read intent and tone, steer the conversation, and still beat some of the best sales teams in the country
. If we get it right, owning a home in the US gets about 1% cheaper. Frontier models aren't built for that. We are.

Outperforming human sales teams on hundreds of thousands of calls a day is not easy. Here's what you'd actually be building.

• Models are trained to be agreeable. Ours has to sell.
• Persuade and adapt to a real person without going off-script
• A research-flavored problem the labs aren’t optimizing for

• What sounds like one agent is a swarm of specialists handing off mid-call
• We rewrite the agent’s context, tools, and instructions between turns
• All inside a few hundred milliseconds, every single turn

• Conversations aren’t functions that return true or false
• Turn-level evals on real moments, plus full end-to-end simulations
• Simulated callers, LLM-as-judge, and metrics per behavior

• People tell our agents genuinely personal things. That is not abstract.
• Security and privacy are design constraints, not a checkbox
• Bounded autonomy — unsafe actions simply don’t exist for the agent

The Team

This might sound simple but it is rigorously enforced. Our employees are the soul of the company and we both demand and reward success.

Uncommonly bright. You can hold a messy problem in your head and make it simpler, approach problems from creative angles, and navigate ambiguous and novel situations with clarity and speed.

Writing, thinking, and exchanging views is at the heart of our decision making process. We judge by the quality of the idea and the evidence, not the pedigree of the person who is sharing it.

Everyone's contributions must be impressive and significant, but they also must be integral to the success of the company. A team victory rewards all.

Studied at

Built things at

Google Meta Apple Amazon

Culture

These three values govern how we work.

We insist on human-level fluency. Every conversation should feel natural, capable, and polished.

Our product has lots of surface area and we expect each of our employees to own a large part of it. Your time is your own; self-starters and problem solvers will be celebrated.

We work hard and we don't take ourselves too seriously. The intensity and velocity are infectious.

Generous where it counts: top-of-market health, early equity, four days in SF, and dinner on us.

Top-of-market health coverage for you and your dependents.

Comprehensive dental and vision coverage for you and your dependents.

A top-of-market employer match on your retirement contributions.

Real ownership in what we’re building, granted at an early stage.

We hire exceptional people to build together in San Francisco, and pay accordingly.

Take the time you need. We mean it — please actually use it.

Snacks around the office, and dinner on us.

Four days a week together in SF, with a flexible remote day on Wednesdays.

Come Work Here

We're hiring people who can own a large piece and still make the whole better. The product is already in market, the problems are hard, and the team is still small enough that you'll change what we build.

Full time

On-site

About Sela

Sela is the market-leading AI voice agent for mortgage lenders. Founded in 2024, we’ve grown like crazy (~$0->13M in 20 months), and currently count 5 of the 10 largest independent mortgage banks in the U.S. as enterprise customers. If we’re successful, the cost of owning a home in the US will become ~1% cheaper. We’ve raised $21M across our Seed and Series A from top investors.

Why this is a huge opportunity

Around 1% of the U.S. labor force works in phone sales, representing…