Research Engineer
| Company | Letta |
|---|---|
| Location | San Francisco, CA, USA |
| Salary | $Not Provided – $Not Provided |
| Type | Full-Time |
| Degrees | |
| Experience Level | Junior, Mid Level |
Description
Our background
Letta is a company founded around the MemGPT project (13k+ GitHub stars). The founding team comes from the same research lab and PhD advisors at Berkeley that produced Spark (→ Databricks) and Ray (→ Anyscale). We have deep expertise in both AI and systems, are currently hiring a founding team of exceptional engineers to join us in building the next generation of LLM agent technology.
👾 You can read more about Letta on TechCrunch and our blog.
LLMs are one piece of a complete agentic system. To build human-like AI that can reason, plan, learn, and remember, we need to engineer the new computer. We believe that the key research advancements that unlock these capabilities will happen from fast-paced open research at the application layer that mixes and matches the best models.
At Letta, our research is focused on understanding the fundamental limitations of LLM-driven intelligence, and building the open model-agnostic application layer above LLMs. As a research engineer, you will work closely with a world-class research team behind MemGPT (PhDs from UC Berkeley’s BAIR and Sky research labs) on agentic systems, memory, reasoning, and scaling test-time compute. You will productionize research developments in Letta’s OSS framework and cloud platform.
Responsibilities:
Developing Letta’s core agentic loops, which encompasses tool execution, stream parsing, reasoning, and more.
Evaluating and improving performance of Letta’s agents framework with open models and new model types (e.g. reasoning models)
Integrating LLM API providers into Letta’s framework
Skills:
Extremely strong programming skills in Python
Strong AI/ML fundamentals
Ability to stay up to date with new developments in agents/memory/LLM space
Experience in software engineering
Examples of what you might work on:
Adding memory management capabilities to voice agents
Implementing constrained decoding to improve tool calling in open models
Building systems for multi-agent coordination
Benchmarking model performance
Our hiring process
We are hiring a small, tight-knit team of exceptionally talented founding engineers. Every hire matters, so we take the hiring process very seriously.
Initial phone interview (30m video call): We want to learn more about your background, your skills, your opinions on open source AI, and why you want to work at an early stage AI startup.
Technical take-home (<1hr assessment): To get a better sense of your skillset, we’ll give you an example problem to work that’s as targeted to your potential day-to-day work as possible.
Paid workday (in-person recommended): As the final step in the interview process, we’ll simulate working together as closely as possible by giving you a real (or as close to real as possible) task to work on for a day – and paying for your time of course. If you live in the Bay Area, we highly recommend visiting our offices in-person! We’re an in-person company, so working at our office will give you a great idea of what it will be like to join as a full-time member of the team.
Benefits
Not Specified