Deskripsi Pekerjaan
At Nexus Intelligence Labs, we are architecting the next generation of autonomous AI agents capable of complex reasoning and tool use. We are looking for a Senior Machine Learning Engineer to join our core research team in the heart of San Francisco. This is a high-impact role where you will own the full lifecycle of foundation models—from data curation and distributed training to optimization and production deployment.
You will work alongside published researchers and world-class engineers to push the boundaries of LLM efficiency and agentic workflows. We offer a compute-rich environment (dedicated H100 clusters), a culture of radical candor, and the autonomy to define the technical roadmap for our flagship products.
Why This Role Is Different
- Compute Freedom: Unblocked access to massive GPU clusters for experimentation.
- Research-to-Production: Your models ship directly to enterprise customers within weeks, not years.
- Ownership: End-to-end ownership of critical model architectures.
Tanggung Jawab
- Design, implement, and scale novel model architectures (Transformers, MoE, RNN alternatives) for agentic AI systems.
- Optimize distributed training pipelines (FSDP, DeepSpeed, Megatron-LM) across 100+ GPU clusters for maximum MFU.
- Lead the development of custom kernels (Triton/CUDA) to accelerate attention mechanisms and inference throughput.
- Build robust evaluation frameworks (automated + human-in-the-loop) for reasoning, coding, and tool-use capabilities.
- Collaborate with Product to define data strategies: curation, synthesis, and RLHF/RLAIF pipelines.
- Mentor junior engineers and establish best practices for MLOps, reproducibility, and model observability.
- Stay at the bleeding edge of ML literature; reproduce and adapt SOTA papers (e.g., mixture-of-depths, speculative decoding, test-time compute).
Kualifikasi
- M.S. or Ph.D. in Computer Science, Machine Learning, or equivalent practical experience (5+ years industry ML).
- Proven track record of training and deploying large-scale generative models (>10B parameters) in production environments.
- Deep expertise in PyTorch/JAX internals, distributed systems (NCCL, Gloo), and GPU architecture optimization.
- Strong proficiency in Python, Rust, or C++; experience with Triton Lang or CUDA is a significant plus.
- Demonstrated ability to debug training instability (loss spikes, gradient divergence) and optimize memory/throughput trade-offs.
- Experience with RLHF, DPO, PPO, or Constitutional AI alignment techniques.
- Excellent communication skills; ability to distill complex research concepts for cross-functional stakeholders.