Role: Principal ML Systems & Alignment Engineer
Location: Dallas, TX (Hybrid – 3 days/week onsite)
Duration: 12 Months
Job Description:
We are seeking a highly experienced Principal ML Systems & Alignment Engineer with deep expertise in Large Language Models (LLMs), Small Language Models (SLMs), distributed model training, agent evaluation, and AI alignment. This is not a conventional Machine Learning Engineer position. The ideal candidate must combine advanced ML research capabilities with strong software and systems engineering skills to architect benchmarking platforms, develop model alignment pipelines, and optimize large-scale training infrastructure.
A Ph.D. in Deep Learning, Machine Learning, Artificial Intelligence, or a closely related quantitative discipline would be a strong plus.
Key Responsibilities:
1. LLM/SLM Agent Benchmarking & Evaluation Platform
- Design and build scalable, deterministic benchmarking platforms to evaluate multi-step LLM/SLM agents and their ability to complete complex tasks.
- Develop stateful, isolated testing environments to simulate real-world tools, APIs, databases, and external system interactions.
- Establish automated evaluation metrics covering task completion, execution efficiency, tool accuracy, error recovery, and agent self-correction.
- Build AI-driven evaluation systems using LLM-as-a-Judge, automated critique mechanisms, verification workflows, and reward-model selection to minimize manual evaluation.
2. Model Training, Fine-Tuning & Alignment
- Lead end-to-end development and fine-tuning of domain-specific LLMs and SLMs using Supervised Fine-Tuning (SFT) and advanced alignment techniques.
- Develop and optimize Reinforcement Learning from AI Feedback (RLAIF) pipelines using methods such as PPO, DPO, GRPO, and IPO.
- Implement reward modeling strategies and tune KL-divergence constraints to align model outputs with enterprise policies and business requirements.
- Integrate model training and alignment workflows with enterprise knowledge systems, context layers, access controls, and security policies.
- Scale distributed model training across multi-node GPU environments using DeepSpeed, FSDP, or Megatron-LM.
3. Advanced ML Research & Custom Model Engineering
- Analyze and translate advanced research papers from ArXiv, NeurIPS, ICML, and other leading ML research venues into practical implementations.
- Develop high-performance custom PyTorch and CUDA components, including neural network layers, loss functions, and tensor operations.
- Modify model architectures and internal components, including attention mechanisms and Mixture-of-Experts (MoE) routing strategies.
- Profile and optimize training and inference workloads to address GPU utilization, gradient computation overhead, memory fragmentation, and compute bottlenecks.
- Collaborate on research-driven improvements to model performance, efficiency, and alignment capabilities.
Required Technical Skills & Qualifications:
- Deep Learning & Frameworks: Expert-level Python and PyTorch proficiency, including custom tensor operations, neural network layers, and loss function development.
- Distributed Training Infrastructure: Hands-on experience with DeepSpeed ZeRO, FSDP, or Megatron-LM across multi-node NVIDIA A100/H100 GPU clusters.
- GPU & Networking: Strong understanding of CUDA, NCCL communication, GPU memory optimization, distributed computing, and cluster-level performance bottlenecks.
- LLM Alignment & Reinforcement Learning: Production-level implementation experience with PPO, DPO, or GRPO, including reward modeling, preference optimization, and KL-divergence tuning.
- Agent Benchmarking: Experience building automated evaluation frameworks for multi-step AI agents, tool usage, task completion, and self-correction.
- Software Engineering: Advanced Python development, object-oriented design, concurrency, asynchronous programming, and scalable system architecture.
- Secure Execution Environments: Experience developing isolated sandbox environments using Docker, gVisor, or equivalent technologies.
- AI Data Engineering: Experience with synthetic data generation, automated prompt optimization/evolution, and preference-pair collection for model training and alignment.
Preferred Qualifications:
- Ph.D. or Master's degree in Computer Science, Artificial Intelligence, Deep Learning, Machine Learning, Mathematics, or a related quantitative discipline.
- Research publications or open-source contributions involving LLM training, model alignment, reinforcement learning, or agent evaluation.
- Experience developing enterprise-grade autonomous AI agents that interact with external tools, APIs, databases, coding environments, or browser automation systems.
- Strong ability to translate advanced ML research into production-ready systems and scalable engineering solutions.
Important Candidate Requirements:
This position requires a combination of advanced ML research, LLM alignment, distributed GPU training, and systems engineering. Candidates with only traditional ML model development, data science, or standard Python-based ML experience will not be a suitable fit.
Candidates should demonstrate hands-on experience in:
- Building LLM/SLM agent benchmarking and evaluation platforms.
- Implementing production-grade PPO, DPO, or GRPO alignment pipelines.
- Scaling distributed training across multi-node GPU clusters.
- Developing custom PyTorch/CUDA components and optimizing model architectures.
- Engineering secure agent execution environments and automated evaluation systems.