9 hours ago
Responsibilities
- Post-train open-weight language models on proprietary legal data from base-model selection through training, evaluation, and iteration.
- Apply supervised fine-tuning, preference optimization, reinforcement learning, reward design, grader design, and related post-training techniques.
- Build training datasets and data pipelines with labeling guidance, model-generated data, and human feedback loops.
- Own reliable training infrastructure and diagnose or optimize distributed training runs.
- Build and operate model-serving systems, agents, evaluation pipelines, and supporting production infrastructure.
- Partner with product and agent engineers on model-versus-agent system design.
- Work with lawyers and domain experts to translate real workflows into model, data, and evaluation decisions.
Requirements
- Hands-on experience with LLM post-training using PyTorch or equivalent frameworks.
- Understanding of training, evaluation, and inference systems, as well as open-weight models and post-training techniques.
- Experience building agents, tools, services, and production infrastructure around models.
- Ownership of model development or post-training work in applied settings that shipped to real users.
- Ability to make pragmatic trade-offs across model research, product engineering, agents, tools, context, and workflows.
- Meaningful ownership of AI, applied research, or adjacent engineering work that materially improved model capability or product outcomes.
Tech Stack
Categories
About Carta
Carta builds software and services for private companies, investors, and law firms to manage cap tables, equity plans, valuations, and fund administration. Its SaaS and administration offerings provide a system of record for founders, VCs, and LPs, and support transactions such as tender offers. Founded in 2012 and headquartered in San Francisco, the privately held company is used by 55,000+ companies and 10,000 funds representing $250B+ in assets under management.