Research Engineer — Gauntlet

The Bitcoin Street Journal; New York, San Francisco, Los Angeles (NY, CA, CA); Hybrid; USD 120,000 - 180,000; Full time

Job summary

Gauntlet is seeking an AI Research Engineer to help improve how we build, evaluate and use AI agents across the business.

We already operate agents in production, including Aria, an in-house generalist agent, used across research, engineering, and business workflows.

Responsibilities

• Build evaluation frameworks to measure AI agent performance on real tasks.
• Collaborate with engineers and researchers to identify weaknesses and design experiments.
• Contribute to existing agents and internal tooling; start with engineering-focused shipping improvements.

Job description

• Company: Gauntlet
• Location: Remote
• Sector: CryptoPerp
• Source: web3.career

Research Engineer

Full Time /

Remote

The role

Gauntlet is looking for an AI Research Engineer to help improve how we build, evaluate and use AI agents across the business.

We already have agents running in production, including Aria, our in-house generalist agent. They're used across research, engineering and business workflows, and we've learned a lot from seeing how they perform in practice. Some tasks work extremely well; others produce inconsistent results, and changes to models or tooling can introduce unexpected problems. Today, much of that feedback is still subjective. We want to build a more systematic way of understanding what works, what doesn't and why.

A major part of this role will be building the evaluation frameworks that allow us to measure agent performance on real tasks. You'll work with our engineers, quantitative researchers and other teams to understand where agents struggle, turn those observations into measurable experiments, and use the results to improve the systems We're building.

You'll also contribute directly to our existing agents and internal tooling. Initially, the work will be more engineering-focused as you get familiar with our systems, shipping improvements and