Latest AI JobsBrowse Jobs
Cognition logo

Research Engineer, ML Infrastructure

CognitionSan FranciscoResearch & Development
Apply on company site
Salary
Not listed
Job type
Full-time
Posted
6 Aug 2026
Status
Open · checked 7d ago

About this role

Cognition is seeking a full-time Research Engineer, ML Infrastructure in San Francisco to build and scale the core systems that power advanced AI software agents. This role is ideal for an experienced systems engineer who wants to work directly alongside researchers to accelerate frontier AI development.

Role Overview

The mission of this role is to ensure that experimental infrastructure never becomes a bottleneck for research velocity. Every training run, evaluation loop, and iteration depends on fast, reliable, and scalable systems. The professional joining this team will own distributed training infrastructure, experiment orchestration, data pipelines, and the tooling that transforms raw compute into efficient research workflows within an applied AI laboratory setting.

Key Responsibilities

Build and manage distributed training infrastructure across GPU clusters, focusing on reliable job launchers, checkpointing, recovery, fault tolerance, and comprehensive monitoring.
Scale the infrastructure responsible for running hundreds of thousands of concurrent coding agent rollouts in virtual machine sandboxes.
Profile and optimize end-to-end training throughput by identifying and resolving bottlenecks in data loading, communication, memory utilization, and compute efficiency.
Design, maintain, and improve experiment orchestration and tooling systems to help researchers launch, track, and analyze experiments efficiently.
Engineer high-throughput, dependable data pipelines that guarantee data quality, reproducibility, and efficiency for large-scale training and evaluation runs.
Diagnose and fix complex training failures spanning GPUs, networking, numerics, and data to ensure graceful failures and fast recovery.
Implement and refine parallelism strategies, including data, tensor, pipeline, and sequence parallelism, to maximize hardware utility.
Anticipate upcoming research requirements and proactively build scalable infrastructure solutions ahead of time.

Requirements

Demonstrated experience building and operating distributed training systems for large models, spanning from the cluster level down to the training loop.
Strong fundamentals in systems engineering, covering distributed systems, networking, storage, and cross-stack hardware-software performance analysis.
Proficiency in Python and C++, alongside hands-on systems-level experience with deep learning frameworks like PyTorch.
Practical background in GPU performance profiling, memory optimization, and compute efficiency enhancement.
Experience implementing and optimizing parallelism strategies for large model training.
Proven track record of developing tooling and abstractions that accelerate research workflows.
Strong debugging instincts for complex, non-deterministic distributed systems.
Sufficient machine learning knowledge to engage substantively with researchers regarding architecture choices and infrastructure demands.

Skills and Tools

Python
C++
PyTorch
Distributed Systems
GPU Performance Profiling
Parallelism Strategies

About the Company

Cognition is an applied artificial intelligence lab based in San Francisco that builds end-to-end software agents and is recognized as the creator of Devin, the first AI software engineer. The organization brings together a talent-dense founding team featuring world-class competitive programmers, former founders, and leaders from cutting-edge AI organizations.

How to Apply

Candidates can apply by visiting the official company careers page on the Cognition website.

Skills and tech stack

PythonPyTorchC++

Questions candidates ask

Where is this job located?

This position is based in San Francisco, California.

What is the primary focus of the Research Engineer role?

The role focuses on building and scaling core ML infrastructure, including distributed training systems, experiment orchestration, data pipelines, and performance optimization for AI research.

What programming languages are required?

Candidates must be proficient in Python and C++.

What kind of systems will I be working with?

You will own infrastructure running across thousands of GPUs, handling distributed training jobs, GPU cluster management, and large-scale agent rollouts in VM sandboxes.

How should I submit my application?

Interested applicants should apply directly on the official company careers page.

Ready to apply?

The application is on the official Cognition careers page.

Apply on company site