---
title: 'Machine Learning Engineer - Pre Training at Mindbeam'
canonical: 'https://feeny.ai/job/machine-learning-engineer-pre-training-mindbeam-united-states-hr7774xvtqwc'
type: 'job'
last_seen: '2026-11-01'
---

# Machine Learning Engineer - Pre Training at Mindbeam

- **Company:** Mindbeam
- **Location:** United States
- **Compensation:** $150k–$190k
- **Employment:** full-time
- **Work type:** remote
- **Posted:** 2026-07-04
- **Last confirmed live:** 2026-11-01
- **Apply:** https://jobs.ashbyhq.com/mindbeam/cc155563-3845-4b4b-9f00-3c31d0105393/application

**Skills:** Python, PyTorch, TensorFlow, JAX, Docker, Kubernetes, GPU scheduling, Memory optimization, Parallelism strategies, High-performance computing

> Design and optimize large-scale pre-training pipelines for foundation models, implementing distributed training strategies and fault-tolerance systems while collaborating with researchers to translate experiments into production-ready workflows.

## Job description

## About Mindbeam

We are building the next-generation AI infrastructure for open source and enterprise. Our work is deeply research-oriented and passionate about developing ground-breaking innovations to take state-of-the-art AI applications to the next level. What drives us is not only advancing technology, but empowering the people behind it. We are a community of researchers, engineers, and visionaries who believe that collaboration, curiosity, and openness fuel progress. If you’re motivated by impact and inspired to build tools that others can build upon, you’ll be in the right place.

Mission Design and optimize large-scale pre-training systems that power Mindbeam’s generative AI models.

## Role Expectations

- Build scalable pre-training pipelines for foundation models, optimizing throughput and efficiency.
- Implement distributed training strategies across GPUs/TPUs and high-performance clusters.
- Collaborate with researchers to translate experimental setups into production-ready workflows.
- Develop monitoring and fault-tolerance systems to ensure reliable large-scale training.
- Continuously benchmark and tune performance across hardware and software stacks.

Background

- Bachelor’s, Master’s, or PhD in Computer Science, Engineering, or related field—or equivalent experience.
- 2+ years of experience with large-scale model training and distributed systems.
- Strong coding skills in Python and familiarity with ML frameworks (PyTorch, TensorFlow, JAX).
- Experience with GPU scheduling, memory optimization, and parallelism strategies.
- Comfort with containerized and orchestrated environments (Docker/Kubernetes).
- Understanding of high-performance computing and networking bottlenecks.

## About You

You thrive on scale and complexity. You enjoy solving system-level bottlenecks, pushing hardware and software to their limits, and working closely with researchers to accelerate cutting-edge AI development.

## About Mindbeam

## Company Overview
- **One-liner**: Mindbeam builds next-generation AI infrastructure, primarily through its Litespark framework that accelerates LLM training and inference while dramatically reducing energy consumption.
- **Entity Type**: Private (Bootstrapped – has not raised any funding)
- **Headquarters**: New York City, United States
- **Founded**: 2024
- **Founders**: Nii Osae (Founder & CEO)

## Core Business
- Primary industry/industries: AI Infrastructure, Machine Learning, LLM Optimization, Generative AI, Enterprise Software, Energy Management
- Target customers: B2B; enterprises and research labs training large language models; organizations looking to reduce GPU costs and energy consumption
- Mission or purpose statement: Building next-generation AI infrastructure (per LinkedIn)

## Products & Services
- **[Litespark]**: High-performance LLM framework that speeds up training and inference while improving GPU efficiency. A drop-in replacement for PyTorch that requires zero code changes. Key metrics: up to 6x higher throughput per GPU, up to 83% lower energy consumption, up to 88% higher MFU for multi-node training. Compatible with NVIDIA GPUs and CPUs (including Apple Silicon M5, AVX-512, Intel Core Ultra).
- **[Litespark-Inference]**: Inference engine that runs large models on standard CPUs without GPUs, achieving 2x faster performance on Apple Silicon M5, AVX-512, and Intel Core Ultra platforms.
- **[SpinGQE]**: A generative quantum eigensolver for spin Hamiltonians (open-source research project on GitHub).

## Market Standing
- **Valuation/Market Cap**: Not applicable (bootstrapped, no funding rounds)
- **Key Metric**: No revenue disclosed; company is unfunded with ~8-9 employees.
- **Notable Investors/Partners**: No investors. Partnerships appear through AWS (spotlighted by AWS Startups, presented at AWS NYC Summit). No formal partnership announcements.
- **Growth Signals**: 
  - Monthly website traffic grew +643.9% (to 2,336 visits) per LinkedIn.
  - LinkedIn followers: 739 as of mid-2026.
  - Active recruiting: multiple open positions listed on Ashby careers page.
  - Technical report released in October 2025 with benchmark results.
  - Active GitHub with three public repositories.
  - Featured by AWS Startups (August 2025).

## Competitive Advantages
- **Performance without code changes**: Litespark is a zero-code PyTorch drop-in – integrates with existing PyTorch and NVIDIA workflows, reducing adoption friction.
- **Dramatic energy and cost savings**: Up to 83% less energy and 6x faster training on same hardware translates directly to lower infrastructure costs.
- **CPU inference capability**: Enables running 2B parameter models on standard CPUs, expanding deployment options.
- **Strong early benchmarks**: Public benchmarks show measurable gains over baseline, which is rare for a young bootstrapped company.

## Strategic Focus
- Continuing to refine and commercialize Litespark for large-scale pre-training and inference.
- Building out the team (hiring across research, engineering, and operations).
- Expanding AWS Marketplace presence (GitHub repo for AWS Marketplace usage instructions).
- Research into quantum computing (SpinGQE) suggests long-term R&D ambition.

## Why Work Here
- **Culture**: Small, early-stage team (~8 employees) with a high density of talent from top organizations (Amazon Web Services, JPMorgan Chase, Cleveland Clinic, UC Berkeley, etc.). Emphasis on breakthrough performance and energy efficiency.
- **Remote/Hybrid/Office**: Not explicitly stated. Headquarters in New York City; likely hybrid given NYC base.
- **Notable Perks**: Opportunity to work on cutting-edge AI infrastructure with significant impact on cost and energy. Founding team access. Open source contributions.
- **Engineering Culture**: Research-heavy (two research scientists on staff), open-source friendly (GitHub), and hands-on with GPU and system optimization.

## Sources
1. [mindbeam.ai](https://mindbeam.ai)
2. [LinkedIn - Mindbeam AI](https://www.linkedin.com/company/mindbeam-ai)
3. [GitHub - Mindbeam-AI](https://github.com/mindbeam-ai/)
4. [Tracxn - Mindbeam](https://tracxn.com/d/companies/mindbeam/__88E92mpJA3ms2zlbSYwBFJQWzM1iaZ4IbxzK6Cz76Ng)
5. [Ashby Careers - Mindbeam](https://jobs.ashbyhq.com/mindbeam)

## Other roles at Mindbeam

- [Solutions Architect - Contractor](https://feeny.ai/job/solutions-architect-contractor-mindbeam-melbourne-3ehbnnwmwc2h) — Melbourne, Australia