Espresso AI is a company using LLMs to build neural optimizers, scheduling systems, and workload tuners, initially focused on making data warehouses and Spark jobs more efficient. They are hiring Staff ML, Staff Infra, and Full-Stack Data Engineers (FDEs) for highly technical, user-facing roles. The core idea is to predict and optimize compute resource usage for complex data workloads.
What they want, where you stand, and the exact résumé edits to qualify.
Biggest lever: Gaining practical experience in training and deploying custom ML models for system optimization, not just consuming LLM APIs.
A starter prompt for Claude Code, what you'll need, and how to reach them.
You are a Staff ML Engineer. Your task is to design and implement a minimal viable product (MVP) for a 'neural optimizer' that predicts compute resource consumption for a simple Spark job based on its configuration and input data size. Focus on the core prediction engine using a basic neural network. Use Python with PyTorch/TensorFlow, scikit-learn for data preprocessing, and FastAPI for a lightweight prediction API. The MVP should accept job parameters (e.g., input data size, number of partitions, Spark executor memory) and output predicted CPU and memory usage, and estimated runtime. Assume synthetic or pre-collected historical data for training; do not worry about real-time telemetry ingestion for this MVP. Define clear data schemas for training and inference. Provide the Python code for the model definition, training loop, and FastAPI endpoint. Include instructions for setting up a virtual environment and running the API. The output should be a single Python file `main.py` and a `requirements.txt`.
Espresso AI | Staff ML, Staff Infra, FDE | Brooklyn or San Francisco | Full time We're using LLMs to build neural optimizers, neural scheduling systems, and neural workload tuners. (If you're ex-Google, you can think of it like Borg powered by LLMs.) Today we use ML to make data warehouses and spark jobs more efficient. We're hiring staff ML engineers to train models that can understand how much compute a job needs, how it scales to larger machines, whether a machine can run more jobs, and so on; and staff infra engineers to take those models and deploy them on real-world production systems. We're also looking for FDEs who can help us talk to users and run pilots. This is a pretty technical role (you need to be able to do data analysis and debug in prod) that's also user-facing - it should be a good fit for a former (or future) technical founder. If this sounds cool, please email me: alexis [at] espresso [dot] ai
Build a small, public project that trains a simple ML model (e.g., a small neural net using PyTorch/TensorFlow) to optimize a resource allocation problem, then deploy it via a simple API or serverless function. Focus on the training loop, data preparation, and a basic deployment strategy. (Estimated time: 2-3 months part-time).
Standard programming language.
New: Deep learning framework for neural networks - ~1-2 weeks for core concepts.
Learn it: Search getting-started ↗
Standard ML library for data preprocessing.
Known for building performant Python APIs.
Significant learning curve for model training, deployment, and monitoring - ~3+ months for production readiness.
Learn it: Search getting-started ↗
Email alexis [at] espresso [dot] ai directly as specified in the hiring post.
“I'm a solo operator with strong AI coding skills and a background in building complex systems. I've prototyped a neural optimizer for Spark job resource prediction (demo available) and am interested in how my skills could contribute to Espresso AI's mission to optimize data warehouses. Happy to discuss the FDE role and specific project needs.”
Open the original ↗