Staff / Principal Software Engineer [Los Angeles/Bay Area] Los Angeles

TubeScience; Los Angeles (CA); Hybrid; USD 180,000 - 250,000; Full time

Job summary

TubeScience is hiring a Staff/Principal Software Engineer to own the architecture direction for two product areas in Los Angeles/Bay Area. You’ll steer end-to-end AI-powered pipelines that transform prototypes into production-ready video and media systems.

Expect direct impact, weekly ship cycles, and collaboration with senior engineers and product managers. You’ll work on platforms that manage model routing, executor agents, and data-driven decisions, with hands-on experience in C/C++, Python,

Qualifications

• Experience shipping AI systems and production-grade pipelines.
• Hands-on with Python and TypeScript in production.
• Built platforms from zero and made foundational calls.
• Experience with vendor APIs, agents or tool calling.
• Experience in adtech, production studios, or similar environments is a plus.

Responsibilities

• Own the architecture direction for your area and release cadence.
• Direct a fleet of agents and ensure accountability for shipped work.
• Collaborate across digital production and media teams to align architecture.
• Ship code or directives weekly and validate outcomes with users.

Skills

C/C++

Rust

Zig

Python

TypeScript

System design

AI in production

Tools

FFmpeg

CDNs

Ingestion tools

Editing tools

Adobe Photoshop

Premiere Pro

Job description

Staff / Principal Software Engineer [Los Angeles/Bay Area]

TubeScience Labs’ mission is to create trusted, scalable, and self-improving AI systems that power the largest performance-based paid social creative video company in the world. TubeScience is Meta's largest creative partner and AppLovin's #1 creative partner, producing 8,000+ original ads every month from a 100,000 sq. ft. Los Angeles studio, backed by a library of 1.6 million+ performance ads and $3B in annual managed ad spend. That is what gives TubeScience one of the richest first‑party creative‑performance datasets anywhere.

Labs turns that data — and the playbook behind billions in spend — into frontier AI tools that actually ship. We are an end‑to‑end native agentic AI lab from research, design, coding, experimentation to delivery. Our pipelines run against real creative, real deadlines, and real budgets every day.

You’ll join a small, AI‑empowered, senior team of engineers and product managers, with direct access to expert users and you will own the whole architecture direction for your area. You’ll be expected to work across all frontier and open‑weight models for every phase of your job from discovery to prototyping, to specs and building, and to evals and deployment.

The role

You’ll own the architecture direction for your area: what gets built, in what order, and why, as well as the release cadence and, just as importantly, what doesn’t get built. We don’t expect you to write code by hand; you’ll direct a fleet of agents. But you remain accountable for every line that ships.

Small senior team. No layers between you and the problem. A direct line from your commits to the people using them, who will tell you immediately if it is wrong.

We weigh directly relevant experience heavily. These roles come with real budgets, real footage and real operators from day one, and there is little time to learn the domain from scratch.

We are hiring engineers for two product areas. Each one sets the technical direction for their area, ships code every week, and decides what does not get built, which is the harder half.

• Digital production: The full‑stack system that turns prototypes into finished videos, shifting the work from human editing toward AI‑assembled variants. You understand editing and generative workflows well enough to know where automation holds up and where it does not. It works with both state‑of‑the‑art AI tools and traditional editing software such as Adobe Photoshop and Premiere Pro.
• Media: The systems behind everything after an ad is made and approved: deploying creative to ad platforms, efficiently allocating spend, running experiments, and measuring and reporting performance against client goals. These are several projects that function as one system, with architecture and standards built to last. You bring hands‑on depth in at least one area of advertising systems (media buying, ad‑platform automation, spend allocation, bidding and auctions, measurement and attribution, ad experimentation, or ad ranking and recommendation) and can extend it to the rest.

• You have held Staff or Principal scope. You have set technical direction across multiple systems or teams, and you can name the calls that were yours and what came of them.
• You bring depth in the domain of the area you apply for. For digital production, video editing, media processing or generative media pipelines. For media effectiveness, advertising, media or performance‑data systems.
• You have built a platform from zero, made the foundational calls, and lived with them. We will ask about the ones you regret, and "none" is the wrong answer.
• You have put AI into production beyond chat. Vendor APIs, agents or tool calling, evaluation, permissions and cost control, running against real users.
• You go a layer or two below the abstraction by instinct. You have worked with at least one systems‑level language like C, C++, Rust or Zig somewhere in your history, and it still informs how you build above it.
• You used Python and TypeScript in production, both ends of the stack, at the level where you have opinions about the trade‑offs rather than familiarity with the syntax.
• You have shipped to real customers and handled what came back.

Preferred qualifications

• You have experience with FFmpeg, CDNs, ingestion or editing tools.
• You dealt with model routing at scale.
• You have spent time inside an adtech company, agency or production studio.

The problems to solve

• Robust media pipelines. Many assets a day through ingest, transform and render.
• Durable workflows. Long‑running jobs that survive restarts, partial failure, and being changed while in flight.
• Model routing and spend. Several vendors, several models, permissions and cost tracked per team. Choosing the right model per job without a human in the loop, and knowing when the cheap one is good enough.
• Agents that make real decisions. Planning systems that commit budget, book people and set schedules, and that operators trust enogh to act on without redoing the work by hand.
• Many projects, one system. Several products that share data, users and infrastructure, and only stay coherent if someone holds the architecture and the standards.
• Shipping in days. Going from a suggestion to production in days rather than quarters. That constrains architecture far more than it sounds like it does.