Senior AI Engineer, Post-Training

Carta•San Francisco, NY

21h

About The Position

Carta is the connected platform and AI-native ecosystem for private capital. Built to replace fragmented tools with a single system of record, Carta brings together the software, services, and legal infrastructure that founders use to manage equity, fund managers use to run administration and reporting, and legal teams use to close transactions. Trusted by 55,000 companies and 1.8M+ equity holders in 160+ countries, and 10,000 funds and SPVs representing $250B+ in assets under management, Carta is transforming how private capital operates. Recognized by Fortune, Forbes, Fast Company, Inc. and Great Places to Work. For more information about our offices and culture, check out our Carta careers page. You’ll join Carta’s ML Engineering team, embedded in Carta Law, our legal tech platform built around autonomous AI agents, specialized legal models, document intelligence and contract workflows. You’ll have end-to-end ownership across model development and applied AI, from post-training and evaluation through model serving and the agents and systems built around those models. You'll work closely with the engineers building the product and bringing these capabilities to users.

Requirements

You have hands-on experience with LLM post-training using PyTorch or equivalent frameworks, and understand the training, evaluation, and inference systems around them.
You are equally comfortable building the product around the model, including agents, tools, services, and production infrastructure.
You can work across model and product engineering problems as needed.
You stay current on open-weight models and post-training techniques.
You have owned model development or post-training work in applied settings and built AI systems around those models that shipped to real users.
You can turn ambiguous product or model problems into tractable technical work, make pragmatic trade-offs across research and engineering, and drive projects from idea through production with minimal guidance.
You have strong judgment on model selection, data, training objectives, and evaluation, and know when training is the right lever versus improving the agent, tools, context, or broader product.
You can make and defend those decisions with data and communicate them clearly across technical and domain teams.
You have done ambitious work in AI, applied research, or adjacent engineering roles, with meaningful ownership of the models or systems you built.
You can point to work that materially improved model capability or product outcomes.
Your experience spans both model-level training work and the product and engineering systems around it, from shaping the technical approach through putting it into production.

Responsibilities

Post-train open-weight language models on proprietary legal data, owning the model development lifecycle end-to-end, from data, objective design, and base-model selection through training, evaluation, and iteration.
Apply the right training techniques for the problem, including supervised fine-tuning, preference optimization, reinforcement learning, and related methods, with careful attention to reward and grader design, model behavior, and evaluation.
Build and improve training datasets and data pipelines, including labeling guidance, model-generated data, and human feedback loops with domain experts.
Own the training stack needed to run experiments reliably, using managed or self-hosted infrastructure as appropriate, and understand distributed training well enough to diagnose and optimize training runs.
Build and operate the systems that take models into production, including model serving, agents, evaluation pipelines, and the surrounding tooling and infrastructure.
Partner with product and agent engineers on model/system co-design, deciding what belongs in the model versus the agent harness, tools, context, and workflow.
Work directly with lawyers and other domain experts to translate real workflows into model, data, and evaluation decisions.