Research Engineer
Tessera LabsBase Salary
$200k - $300k/yr
Responsibilities
- Build and scale post-training infrastructure for SFT, preference optimization, and reinforcement learning involving long-horizon tool use and enterprise transformations.
- Develop memory, context, ontology, and knowledge-graph systems for long-running agents.
- Create data-generation and curation pipelines for synthetic landscapes, transformation traces, tool-call trajectories, and curriculum infrastructure.
- Build sandboxed RL environments and execution-and-verification harnesses with automatically scored outcomes.
- Own offline evaluation infrastructure for trajectory-level agent behavior, reproducible task suites, and long-term result tracking.
- Run experiments end to end, including design, execution, debugging, analysis, and interpretation.
- Optimize distributed training and inference throughput through kernels, parallelism, memory, batching, serving, and long-context techniques.
- Take trained models into production serving through quantization, serving configuration, and rollback planning.
- Establish standards for experiment reproducibility, tracking, and result hygiene.
Requirements
- Significant experience training, fine-tuning, or post-training language models with demonstrated results owned by the candidate.
- RL experience such as RLHF, RLAIF, RLVR, GRPO-family methods, or agentic RL is close to a requirement.
- Experience or strong understanding of memory and context for long-running agents through architecture, retrieval, or training.
- Strong software engineering fundamentals and the ability to write experiment code that others can run.
- Fluency in Python and PyTorch or JAX, with the ability to debug distributed training.
- Experience with GPU infrastructure at scale and understanding of training time and memory usage.
- Ability to design, run, and interpret rigorous experiments and distinguish effects, noise, and bugs.
- Clear written communication and interest in taking research results into production.
- Preferred experience includes RL environments, execution sandboxes, verifiable-reward task suites, long-context modeling, knowledge graphs, ontologies, semantic layers, production agent memory, code models, or open-source ML systems.
- Preferred evidence includes contributions to systems such as vLLM, SGLang, PyTorch, Triton, DeepSpeed, Ray, Megatron, or TRL; publications, technical reports, or open-source releases; and an advanced degree in a quantitative field or equivalent industry research experience.
Categories
About Tessera Labs
Enterprise transformations shouldn't take years or cost fortunes. At Tessera, we've built a multi-agent AI platform that cuts ERP transformation timelines from years to weeks and reduces costs by more than half—while delivering first-time-right outcomes with enterprise-grade security and governance. We combine the dependability of a trusted integrator with the speed of an AI innovator. Our vendor-agnostic platform is pre-trained on thousands of enterprise landscapes and hundreds of years of expertise, so it adapts to your environment from day one—harmonizing systems and data to deliver secure, governed execution across ERP and other critical systems. Unlike traditional system integrators, ERP vendor tools, AI point solutions, or in-house builds, Tessera delivers transformations that compress timelines by 90%, replace people-heavy manual work with intelligent automation, ensure systems and data harmonize correctly from the start, and avoid vendor lock-in through adaptive intelligence that evolves with your workflows. Tessera creates lasting change in how your processes and systems operate together. As the platform adapts to your environment and evolves with new data, you gain the confidence to modernize with less risk—freeing up resources to reinvest in growth rather than maintaining legacy complexity. Proven in regulated industries. Built for enterprise reality. Reshaping how organizations transform.