Applied Scientist

The Role

We are looking for an Applied Scientist to own the models that decide which ad appears in which AI conversation, in real time, across millions of queries a day. Ad selection inside LLM chats is a genuinely new problem: no click history, no cookies, no page context. Just an evolving conversation and a few hundred milliseconds to make a decision that has to feel native to the user, relevant to the advertiser, and profitable for the publisher.

This is a research role with a production mandate. Your models go live, not into a paper. You will work directly with our CTO and the founding engineering team on the systems powering millions of ads a day across Fortune 500 campaigns.

Core Tech Stack

• ML: PyTorch, embedding-based retrieval, modern transformer architectures

• Backend: Python (FastAPI), async serving, real-time inference

• Data: PostgreSQL for transactional, ClickHouse for analytics and event data

• Infra: AWS, Docker, CI/CD

• Plus: whatever the problem demands. We pick tools that ship.

Example Projects

• Design and train contextual ad relevance models for LLM conversations, from retrieval through ranking through creative selection.

• Build efficient algorithms for real-time bidding and auction mechanics, balancing publisher yield and advertiser ROAS.

• Productionize regression models to predict downstream conversion from early conversational signals.

• Apply privacy-preserving clustering methods to categorize conversational context and improve advertiser targeting without compromising user data.

• Run live experiments against real traffic on our publisher network. Iterate on CTR, ROAS, and conversation quality metrics.

• Own the evaluation framework for ad quality, relevance, and user experience across the marketplace.

• Read an ArXiv paper on Monday. Have a prototype running on Friday.

What You Will Do

• Own the research and shipping of the models powering ad selection across Thrad's network.

• Move research to production end to end: hypothesis, training, evaluation, deployment, monitoring.

• Collaborate directly with the CTO and co-founders to translate research direction into working systems, often within the same day.

• Partner with the Data and Product teams to design and analyze live experiments on real traffic.

• Contribute to system architecture decisions as we scale from millions to billions of daily ad impressions.

• Communicate findings clearly to both technical and non-technical audiences across the team.

Who You Are

Must-haves

• 2 to 5 years of applied ML experience, ideally in ads, search, recommendations, ranking, or NLP.

• An advanced degree in Computer Science, Mathematics, Statistics, Physics, Engineering, or a related field — or equivalent experience shipping production ML.

• Strong fundamentals in mathematical and statistical modeling. You know why your model works, not just that it does.

• Fluent in PyTorch, modern transformer architectures, and embedding-based retrieval.

• Strong Python, comfort with distributed training and inference.

• You have shipped ML to production at scale. Not just notebooks.

• You enjoy identifying and owning hard, open-ended problems. You form testable hypotheses and run them to ground.

• Based in San Francisco or willing to relocate. This is an in-person role at our SF HQ.

Nice-to-haves

• Experience with RLHF, preference modeling, or LLM fine-tuning.

• Background in auction theory, bidding systems, real-time pricing, or yield optimization.

• Prior work at an ad-tech, search, recommendations, or marketplace company.

• Published research at NeurIPS, ICML, ACL, KDD, RecSys, or equivalent.

• Familiarity with ClickHouse or other column-oriented OLAP databases for analytical workloads.

• A relentless interest in the AI and LLM ecosystem and how advertising will evolve inside conversational interfaces.

Why Thrad

• First mover: We are defining a new category. Paid ads in AI is an inevitable market and we are building the infrastructure for it.

• Real traction: Fortune 500 clients, millions of daily impressions, and publisher partners across the AI web, all pre-seed.

• Research with teeth: Your work ships. Your models run in production against real traffic with real revenue on the line.

• Direct impact: Small founding team. Your decisions shape the models, the product, and the company.

• Ownership: Meaningful equity in a company at the intersection of two massive markets: AI and ad tech.

• Craft: A team that cares about shipping quality, not just quantity.

How to Apply

Send your CV and a brief note on what excites you about building at the intersection of AI and advertising. If you have published work, a GitHub, or models you have shipped that you are proud of, include them. We read every application.