Founding AI Engineer

Discernis, Inc.; New York (NY); Presencial; USD 180.000 - 240.000; Jornada completa

Ventajas ofrecidas por este puesto de trabajo

Health insurance

Unlimited PTO

Meaningful equity with 4-year vest and

Unlimited PTO plus standard US federal

Descripción de la vacante

Discernis is hiring to take over the AI stack for our legal-discovery platform. You will own agentic workflows, retrieval and evaluation, inference optimization, fine-tuning, and post-training.

In your first ~90 days you’ll work with the founder, then shape our AI direction and build the team around it. We are a seed-stage, NYC-based team shipping to AmLaw firms and Fortune 500 companies, with air-gapped on-prem deployments and private cloud.

Formación

• You’ve trained or post-trained real models (fine-tuning, RLHF, or RL).
• You’ve also shipped LLM or agentic systems to production that real users depend on.
• You have strong evaluation instincts and know how to tell real gains from noise.
• You have hands-on inference and serving experience.

Responsabilidades

• High-recall retrieval and synthesis across billion-token document collections.
• Agents that are reliable, testable, and deployable, not just impressive in a demo.
• Getting strong performance from small open-source models in air-gapped environments.
• Evals that capture what “correct” means to a litigator.
• RL that trades off tool use against compute.

Conocimientos

Model training

RLHF/RL methods

Production deployment

Herramientas

Open-weight models

Private cloud deployment

On-prem GPUs

Descripción del empleo

About Discernis

Discernis builds AI for legal discovery at a scale where general legal AI tools fall apart. When a litigation team needs to know what a group of people knew, and when, across millions of emails, chats, and documents, our platform finds it with the recall attorneys can stake a case on. We deploy in private cloud and fully air-gapped on-prem for AmLaw firms and corporate legal teams.

We’re seed-stage and venture-funded, have paying Big Law customers, and have a pipeline of AmLaw firms and Fortune 500 companies lined up for paid pilots and future deals. Industry incumbents are coming to us to integrate. Discovery is our wedge: the bigger goal is helping enterprises understand and act on their largest, messiest datasets. We’re a small team based in NYC, and you’d be joining before our next stage of growth.

How we work

We value ownership, curiosity, and shipping. We hire people who teach themselves new things, make good calls without much direction, and care about getting it right for the attorneys who depend on us. We work hard at a sustainable pace, without crunch culture or arbitrary deadlines.

The role

Today our founder runs the AI stack personally. He holds a Columbia PhD in AI and built Amazon’s first customer-facing RLHF pipeline. You’ll take the stack over, all of it: agentic workflows, retrieval and evaluation, inference optimization, fine-tuning, and post-training. For your first ~90 days you’ll work side by side with him. After that, you set our AI direction and grow the team around it.

At a larger company, this would be three separate roles. Here, it’s one. This is likely the single highest-impact role at the company, and we need someone who can do the engineering and the science.

Rough split

• ~40% agentic systems: orchestration, tool use, and reliability

• ~20% inference: serving and optimization, including small models on constrained on-prem hardware

• ~40% longer-horizon research: model adaptation, post-training, and RL

Problems you’ll own

• High-recall retrieval and synthesis across billion-token document collections

• Agents that are reliable, testable, and deployable, not just impressive in a demo

• Getting strong performance from small open-source models in air-gapped environments

• Evals that capture what “correct” means to a litigator

• RL that trades off tool use against compute

Our stack

• Models: open-weight models we fine-tune, adapt, and post-train for legal reasoning, then serve as our own versions

• Serving: vLLM on our own H100 and H200 GPUs

• Deployment: the same models run in private cloud and fully air-gapped customer environments

• Language: Python

• Around it: PostgreSQL, Elasticsearch, Kubernetes (K3s), Vue

What we’re looking for

• You’ve trained or post-trained real models (fine-tuning, RLHF, or RL). Experience calling APIs is not enough.

• You’ve also shipped LLM or agentic systems to production that real users depend on.

• You have strong evaluation instincts and know how to tell real gains from noise.

• You have hands-on inference and serving experience.

We’re hiring for this role across levels, from strong AI engineer to AI lead. Compensation reflects the level we hire at.

Nice to have

• Experience on a big tech AI team or at an AI startup

• A PhD or MS in ML

• Research publications or open-source work in LLMs, agents, RL, or evaluation

• Distributed training or large-scale GPU experimentation

Our process

1. A call with our founder

2. A screen with our Head of Engineering

3. A deep dive on a real problem

Why Discernis

• Research problems with a direct path to production and paying customers

• A founder who speaks your language

• Meaningful equity with a 4-year vest and 1-year cliff

• Health insurance

• Unlimited PTO plus standard US holidays

If you need an accommodation during the hiring process, email us at [e-mail].