Machine Learning Engineer - SF Bay Area (Hybrid)

The Opportunity

Secludy is a San Francisco-based, venture capital backed startup driven by a mission to make privacy-preserving AI the industry default and ensure individual privacy withstands the rise of AI.

Our platform enables organizations to securely leverage sensitive data through advanced synthetic data generation with provable privacy guarantees. Our fintech, healthcare, and life sciences customers trust Secludy to protect their sensitive personal data and IP while training AI models.

You will be the first engineering hire at a seed-stage startup solving the "Data Bottleneck" for regulated industries. We are building a proprietary generative engine that synthesizes complex, mixed-type sequential data (fintech, healthcare, insurance) with mathematical privacy guarantees. You aren't just cleaning up scripts; you are building the infrastructure that allows us to iterate 10x faster and deliver to enterprise customers with the highest standard.

What You Will Actually Do

1. Accelerate the Research Loop

* Partner directly with the founders to turn mathematical hypotheses into testable code.

* Refactor for optionality: We need reusable modules for future experiment and re-usability. Instead of rewriting every single line of code, you achieve this by forcing yourself/AI tools to solve problems within strict boundaries, questioning every new file, blocking unnecessary libraries, and demanding that performance gains come from simpler architecture, not spaghetti patches.

* Investigate failures that span the boundary of infrastructure (OOMs, race conditions) and math (loss divergence, gradient explosion).

* Build the harness that ensures when we change the model, we know exactly what happened to our metrics.

2. Productionize the Engine

* Scale Without Sprawl: Orchestrate distributed workloads (parallel training & generation) across GPU clusters using K8s, designing minimal, self-healing deployment architectures that handle scale without creating maintenance nightmares.

* Own the deployment pipeline.

* Implement logging that tells us why a generation job failed, not just that it failed.

* Who You Are

* You work across infra and modeling. You can debug latency and Docker issues, and you can read loss curves and diagnose overfitting.

* You take ownership of broken training runs. You profile GPUs, inspect dataloaders, and find the bottleneck fast.

* You stay pragmatic. You write clean abstractions when they pay off, and you ship a small script when we need an answer today.

Requirements

* 3-6 years of experience in ML Engineering or Applied Research.

* Strong proficiency in PyTorch (you can write a custom training loop from scratch).

* Experience with Generative Models (Diffusion, VAEs, Transformers) beyond just inference.

* Track record of taking experimental code and shipping it to production.

* Comfortable working with Tabular, Free Text, or Time-Series data.

Interview Process (No LeetCode)

We respect your time. Our process tests for the actual work you'll do.

* Screen (30 min): Align on mission and technical background.

* Technical Deep Dive (60 min): Discussion on system design for ML, debugging war stories, and generative concepts.

* Working Session (90 min): We give you a broken piece of our actual research code. You use your own environment/IDE/AI tools to diagnose and fix it while we collaborate.

* Founder Chat (30 min): Final conversation and vision alignment.