Latest AI JobsBrowse Jobs
Letta logo

Research Engineer / Scientist, Post-Training

LettaSan Francisco OfficeTechnical Staff - Research
Apply on company site
Salary
Not listed
Job type
Full-time
Posted
4 Feb 2025
Status
Open · checked 7d ago

About this role

Role Overview

The Research Engineer / Scientist position for Post-Training at Letta is a full-time, in-person role based in downtown San Francisco, requiring five days a week on-site. This opportunity suits ambitious AI researchers and engineers who want to build self-improving artificial intelligence agents that learn from experience and adapt over time.

Key Responsibilities

Pioneer advanced post-training techniques to seamlessly integrate large language models into complete agentic systems.
Train models specifically for improved agentic tool-use, with a strong focus on effective context management.
Design mechanisms for continuous model weight updates after deployment while avoiding catastrophic forgetting.
Design and execute rigorous experiments to understand the relationships between data mixtures, training algorithms, and models.
Build robust infrastructure dedicated to generating and collecting synthetic data at scale.
Create challenging evaluations to measure advanced agentic capabilities.
Share research breakthroughs with the broader community through published papers, technical reports, blog posts, and open-source code.

Requirements

Strong proficiency in Python and deep learning frameworks such as PyTorch.
Deep expertise in post-training techniques, including supervised fine-tuning, reinforcement learning, reward models, and preference learning.
Demonstrated ability to balance execution speed with high empirical rigor.
Proven track record of impactful research through breakthrough publications or significant open-source contributions.
Experience delivering real-world impact that extends beyond purely academic environments.

Skills and Tools

Python
PyTorch
Spark

About the Company

Letta originated from UC Berkeley’s Sky Computing Lab, which is also the birthplace of Spark and Ray. Founded by the creators of MemGPT, the company is backed by industry pioneers like Jeff Dean and Clem Delangue. Letta agents currently power production systems for companies such as 11x and Bilt Rewards. The organization operates as a flat company where every researcher and engineer acts as an individual contributor, valuing execution velocity and a passion for open, accessible AI.

How to Apply

Interested candidates should visit the official company careers page to submit an application and begin the interview process.

Skills and tech stack

PythonPyTorchSpark

Questions candidates ask

Where is this job located?

This role is strictly in-person, operating 5 days a week at the Letta office in downtown San Francisco.

What technical skills are required?

Candidates must be proficient in Python and deep learning frameworks like PyTorch, alongside expertise in post-training techniques such as SFT fine-tuning, reinforcement learning, and reward models.

What will I be responsible for building?

You will train models for agentic tool-use and context management, build synthetic data generation infrastructure, design continuous model update mechanisms, and create agentic evaluations.

What is the interview process like?

The interview process includes an initial 30-minute screen, a 1 to 1.5-hour technical screen, and a 2-day paid in-person work trial onsite in San Francisco.

Is remote work or a hybrid schedule allowed?

No, this is an in-person position with no hybrid or remote options available.

Ready to apply?

The application is on the official Letta careers page.

Apply on company site