Senior Machine Learning Engineer

Integra Tech Advisors; San Francisco (CA); On-site; USD 150,000 - 210,000; Full time

Job summary

Integra Tech Advisors seeks a Senior Machine Learning Engineer to work on LLM interpretability and systems. You will operate inside the model stack, handling weights, activations, and architectures to power a deterministic policy engine for enterprise use.

This hands-on role requires deep knowledge of Transformer models, PyTorch, and the math of deep learning, with a focus on production-grade improvements and scalable deployment across clouds.

Qualifications

• Experience with Transformer architectures, PyTorch internals and the math underpinning deep learning.
• Ability to train, fine-tune or optimize models beyond superficial augmentation.
• Ability to read papers, extract actionable ideas and implement them.

Responsibilities

• Turn mechanistic interpretability research into production code.
• Work with model internals to improve behavior and performance across models.
• Apply techniques like activation patching and control vectors for deterministic outputs.
• Build evaluation and deployment loops to ship changes into enterprise environments.
• Design and optimize feature-level intervention systems for real-time policy enforcement.

Skills

Transformer architectures

PyTorch internals

Deep learning math

Tools

Python

Rust

Node/TypeScript

React

Postgres

Vector databases

Graph databases

Docker

Kubernetes

Terraform

Job description

Senior Machine Learning Engineer — LLM Interpretability & Systems

A venture-backed AI governance startup — born out of Stanford research, backed by institutional investors including Google’s Gradient Ventures, General Catalyst, and Y Combinator — is building the deterministic control layer that lets Fortune 500 companies deploy generative AI with confidence. The platform enforces policy, prevents drift, and produces auditable decisions in real time, bridging the gap between raw LLM capability and enterprise-grade reliability.

About the Role

This is deep, hands-on model work — not prompt engineering. You’ll operate inside the model stack itself, working directly with weights, activations, and architectures to power the company’s Policy Engine, which gives enterprises real-time, auditable control over model behavior in production. The mandate: figure out how a model can be improved for a specific purpose, then build the systems that operationalize that at scale. You’re not just using models — you’re probing the mechanics of how they think.

What You’ll Do

• Turn mechanistic interpretability research into production code
• Work directly with model internals to improve behavior and performance across commercial and open-source models
• Apply techniques like activation patching, control vectors, and feature extraction for targeted, repeatable output improvements
• Build the evaluation and deployment loops needed to ship changes reliably into enterprise environments
• Design and optimize feature-level intervention systems enabling deterministic policy enforcement at inference time

Who You Are

• Strong grasp of Transformer architectures, PyTorch internals, and the math underlying deep learning
• Have trained, fine-tuned, or optimized models beyond surface-level augmentation
• Can read a paper, decide what matters, and implement it
• Takes ownership when something isn’t working
• Motivated by making LLMs reliable and controllable enough for the highest-stakes enterprise use cases

Stack

Python, Rust, Node/TypeScript, React · Postgres, vector and graph databases · Docker, Kubernetes, Terraform across multiple clouds and customer VPCs · self-hosted models across GPU providers plus frontier APIs