Principal Applied Scientist, Small Language Models

Nextdeavor•New York, NY

4h•Hybrid

About The Position

This is a full-time role for a Principal Applied Scientist, Small Language Models in New York City, NY. In this hands-on senior individual contributor role, you will own the small, purpose-built language model that enforces compliance in real time, ensuring messages are safe before they are sent. You will drive model design, evaluation, and production performance, working directly with the founder and cross-functional stakeholders. The role involves a hybrid work model with 3 days onsite in New York City.

Requirements

8+ years in applied ML or AI with models shipped and maintained in production
Hands-on expertise in modern NLP and LLM techniques, including fine-tuning, distillation, and rigorous evaluation
Strong judgment as a senior individual contributor who sets technical direction
Deep instincts for model design tradeoffs (when to fine-tune vs. prompt, when to distill, when smaller models suffice)
Bias for shipping and measuring everything you deploy

Nice To Haves

Experience transitioning systems from large general models to smaller specialized models for cost and latency
Background in compliance, regtech, or other regulated domains
Experience adapting one model to many customers (adapters, per-tenant tuning)
Public research, open source contributions, or technical writing

Responsibilities

Own the compliance detection and rewrite model end-to-end: data, architecture, training, evaluation, and production
Build the evaluation harness and labeled-data flywheel to prove model quality
Optimize and balance accuracy, cost, and latency for production use
Define and execute the path from frontier models to a smaller specialized model that matches or beats accuracy at lower cost and latency
Build infrastructure for per-customer adaptation, including adapters and per-tenant tuning
Translate plain-language policy into enforceable model behavior and document reasoning for model decisions
Run focused error analysis on critical cases (missed violations and false positives)