Founding AI Research Lead - Agentic AI Lab

Fabrion; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Fabrion is designing the future of enterprise AI infrastructure, grounded in agents, knowledge graphs, and multi-tenant governance. We are running research inside the Agentic AI Lab to train and evaluate specialized models for mission-critical enterprise work, with production data access and a benchmark-first plan.

This role covers problem formulation through deployment, with ownership of the research agenda and close collaboration with the founders and platform team.

Qualifications

• PhD in machine learning or closely related field, or an equivalent research record.
• Hands-on experience training sequence models, owning tokenizer, training loop, and evaluation.
• Proven track record shipping research into product or benchmark results.
• Preferred tech stack: PyTorch, Hugging Face, experiment tracking, modern cloud data warehouses.

Responsibilities

• Own the research agenda: model and training design, evaluation protocol, and the publication plan.
• Take models from public benchmarks to live customer shadow deployments with defined gates.
• Set the benchmark discipline with strong baselines and published comparables.
• Lead and grow a small team and collaborate with founders and platform team.
• Write technical plans and papers when results warrant it.

Skills

Research leadership

Sequence modeling

Reinforcement learning

PhD in ML or related

Education

PhD in machine learning

Equivalent research record

Tools

PyTorch

Hugging Face

Experiment tracking

Cloud data warehouses

Job description

Founding AI Research Lead - Agentic AI Lab

San Francisco Bay Area | Full time

Backed by 8VC, we are building a world-class team to tackle one of industry's most critical problems: trusted AI for enterprise operations.

About the Role

Fabrion is designing the future of enterprise AI infrastructure, grounded in agents, knowledge graphs, and multi-tenant governance. We are working on research inside the Agentic AI Lab to train and evaluate specialized models for mission-critical enterprise work.

The direction is specific and ambitious. We share the full thesis under NDA during the interview process. What we can say here: the program has committed design partners with production data access, dedicated compute, a benchmark-first plan with clear go and no-go gates, and a platform team that has already built the governance and serving layer your models will run behind.

This is full-cycle research: problem formulation, data, training, evaluation, and deployment, with your name on the results.

Core Responsibilities

• Own the research agenda: model and training design, evaluation protocol, and the publication plan

• Take models from public benchmark results to live customer shadow deployments, with gates you define and defend

• Set the benchmark discipline: strong baselines first, published comparables cited, results that survive scrutiny

• Lead and grow a small team (ML engineer, data engineer, contractors) and pair closely with the founders and platform team

• Write technical plans internally and papers externally when results warrant it

Desired Experience

• Hands-on experience training sequence models, owning the tokenizer, the training loop, and the evaluation, not only fine-tuning through APIs

• Strong background in at least two of: reinforcement learning (especially offline and imitation settings), sequence decision modeling, structured or constrained generation, learning from event and log data

• A track record of shipping research into a product or landing a rigorous benchmark result

• PhD in machine learning or a closely related field, or an equivalent research record

• Preferred Tech Stack

• PyTorch, the Hugging Face ecosystem, experiment tracking and reproducible training pipelines, modern cloud data warehouses, evaluation harness engineering

Soft Skills & Mindset

• Comfortable as the most senior researcher in the room: setting direction under ambiguity and writing decisions down

• Rigor over hype: you distrust your own results until the baselines agree

• A teacher's instict: part of this role is turning strong engineers into researchers

Why This Role Matters

We believe specialized models built on governed enterprise data can run real, multi-billion-dollar workflows. Your work will not be buried in research reports. It will be benchmarked in public, deployed to real customers, and activated by hundreds of thousands of decisions.