Researcher, Post Training

Owns Lovable’s end-to-end post-training pipeline for language models, adapting reinforcement learning and preference optimization to code-generation and agent workloads. The role combines production engineering, distributed training, evaluation, deployment, and rapid experimentation.

About the job

Responsibilities

Own the full lifecycle of the post-training pipeline, from data curation and training runs through evaluation and deployment.
Apply and adapt reinforcement learning, preference optimization, and supervised fine-tuning methods to improve models for code generation, user-intent reasoning, and reliable agent behavior.
Build evaluation and experimentation infrastructure covering helpfulness, safety, latency, and reliability.
Develop and operate production systems for large-scale training jobs, including GPU orchestration and data pipelines.
Collaborate with agent, product, and infrastructure engineers to turn model gains into product improvements.
Investigate and resolve failures end to end, including training recipes, data issues, and serving regressions.
Read research papers, run experiments, and move promising research into production quickly.

Requirements

Hands-on experience running post-training jobs on large language models, including RFT/RLVR, preference optimization, or similar methods.
Strong production software engineering skills.
Fluency in at least one major ML framework, such as PyTorch or JAX.
Experience with distributed training setups and GPU clusters.
Understanding of the mathematics behind preference optimization, reward modeling, and alignment techniques.
Experience building evaluation systems that measure real-world quality beyond benchmark scores.
Ability to trace model-quality regressions from user-facing symptoms through serving, inference, and training.
Strong execution and focus on shipping improvements to users.

Nice-to-haves

Experience with code-generation or agentic use cases.
Experience putting post-trained models into the hands of real users at scale.
Ownership of the full loop: data curation, training, evaluation, deployment, and production monitoring.
Experience rapidly prototyping ideas from research papers.
Experience with speculative decoding or similar model-efficiency techniques.
Strong opinions on evaluation methodology and experience building evaluations that predict user satisfaction.
Meaningful contributions to the open-source ML ecosystem.

Skills

PyTorch, JAX, Reinforcement Learning, Preference Optimization, Reward Modeling, Supervised Fine-Tuning, Distributed Training, Gpu Clusters, LLMs, Model Evaluation, Data Pipelines, Gpu Orchestration, Speculative Decoding, Code Generation, Agentic Systems