Member of Technical Staff, LLM Post-Training, Applied

About the Role

Sanas is looking for a Member of Technical Staff to lead the post-training and deployment of large language models across a new generation of self-hosted, sovereign-deployed products. This is a rare chance to own applied post-training work end-to-end for text workloads. This role sits at the center of taking strong open-source LLMs and adapting them — through fine-tuning, alignment, and inference optimization — into models that perform reliably in high-stakes, real-world, on-premise environments.

You'll be the technical bridge between what customers need and what actually ships. That means owning engagements end to end — scoping, adaptation, evaluation — and having full say over how text models get shaped and deployed. In between, you'll build the reusable tooling and workflows that make the next engagement faster than the last.

If you care about data quality, evaluation design, and making language models genuinely work in production, this is the role for you.

What You'll Do

• Lead efforts in instruction tuning, preference tuning, and model alignment to ensure models are helpful, safe, and performant in real-world applications.
• Own customer post-training projects end-to-end — from requirements through data generation, training, evaluation, and delivery.
• Customize open-source models for specific customer and product needs, ensuring a seamless path from post-training to serving production workloads.
• Improve inference-time efficiency, reliability, and robustness for high-stakes, real-world deployments — making models dramatically faster and cheaper to run while improving their capabilities.
• Provide technical mentorship and guidance to the team, fostering a culture of engineering excellence and rapid innovation.

What We're Looking For

We need someone who:

• Owns outcomes, not just tasks — carries customer post-training projects from first requirement to final delivery and evaluation, no handoffs mid-stream.
• Sees the whole pipeline as one system — data generation, instruction tuning, alignment, and evaluation aren't separate steps to them, they're one loop that has to work together.
• Cares about what ships, not what publishes — optimizes for model quality and customer outcomes over papers or theory for their own sake.
• Translates in both directions — turns customer needs into technical decisions internal teams can act on, and pushes back when the ask doesn't hold up.

Requirements

Must-have:

• 2+ years of experience building and deploying machine learning-based services in a production environment
• Hands-on experience with data generation and evaluation for LLM post-training
• Experience training or fine-tuning models using SFT, instruction tuning, RLHF, DPO, or similar preference alignment methods
• Strong intuition for text data quality and evaluation design
• Experience with text-specific post-training workflows: chat model alignment, instruction tuning, or text data curation at scale
• Proficiency with the open-source ML ecosystem (Hugging Face, PyTorch) and modern model architectures

Nice-to-have:

• Experience optimizing inference for reduced latency and higher concurrency, especially for on-premise deployments
• Experience improving system performance, efficiency, and scalability of deployed models and applications
• Experience serving low-precision (FP4/FP8) models, multiple LoRA adapters within one model instance (Multi-LoRA), or models distributed across several GPU nodes
• Experience developing large-scale, high-load production systems
• Experience maintaining or contributing to open-source ML projects
• Experience managing machine learning workloads on Kubernetes clusters
• Experience delivering applied ML work to external customers with measurable outcomes
• Familiarity with inference optimization frameworks (vLLM, SGLang, TensorRT)
• Experience building reusable ML tooling or evaluation infrastructure