Founding Full Stack Engineer

Founding AI Engineer – Oncology AI Platform (SF, In-Person)

Why This Role Is Unique

🎯 High-impact, early seat – As hire #7, you'll help architect the intelligence behind a first-of-its-kind AI platform focused on transforming a deeply complex domain of healthcare. You’ll shape product and engineering culture from the ground up.
📊 Unprecedented data advantage – We work with a proprietary dataset comprising 400M+ radiology, pathology, genomic, and clinical records from leading U.S. institutions.
🤖 Frontier-model access – You'll work directly with some of the most advanced model architectures available today, including pre-release access to models from top AI labs, integrated into a highly specialized medical-AI framework.
💰 Massive market, urgent mission – The opportunity sits at the intersection of a multi-trillion-dollar industry and a global health imperative. You'll have the technical canvas to make a significant dent in both.
🧠 Elite, execution-driven team – Collaborate with accomplished founders (including a national champion athlete and ex-intelligence CTO), world-class engineers, and a board of top oncologists from institutions like Johns Hopkins, UCLA, and MSK.

What You’ll Build (First 3 Months)

Architect and deploy multi-agent AI workflows that synthesize structured and unstructured data to produce end-to-end treatment insights.
Own post-training pipelines for cutting-edge LLMs (LoRA, RLHF, RLAIF), tailoring frontier models with medical expertise.
Develop reward models focused on safety, factuality, and outcome-relevance in oncology.
Prototype temporality-aware embeddings to enable longitudinal reasoning across patient timelines.
Expand a bespoke evaluation system with synthetic vignettes, offline policy metrics, and adversarial probing.
Integrate deterministic tools and perception models into reasoning chains (e.g., calculators, image recognition, structured decision aids).
Build and refine retrieval and verification scaffolding, including KAG/RAG modules and factuality-checker agents.

Tough Problems You'll Tackle

Designing agentic planners that chain together reasoning trees across planning, execution, retrieval, and decision support.
Training reward models where signal is sparse, noisy, or contradictory.
Encoding real-world EHR timelines in a way that preserves nuance and disease progression over time.
Creating self-improving systems that turn system failures into new, high-quality training data—without relying on human labeling.

Who You Are

2+ years of experience building from 0→1 with strong full-stack fluency (backend and frontend).
Comfortable in ambiguity, motivated by technical ownership, and driven by real-world impact