About us
Different is a marketing system that runs demand generation for B2B companies. It handles campaign creation, personalization, and optimization in one place, so lean teams can do more without adding headcount, agencies or point AI solutions. Customers use Different to launch campaigns faster, reach the right buyers, and continuously improve marketing ROI. Between them, Different’s founders have founded two unicorns, led AI at a global marketing software giant, and built platforms used by millions. They bring deep product, technical, and business experience to one shared goal: reimagining how modern marketing works. We are backed by General Catalyst and Lightspeed.

The role

You'll build the agents, evals, and model systems behind Different.
This is not a foundation model research role. The job is to take fast-moving models and make them useful in a real product. These systems need to be reliable enough to act on a marketer's behalf and practical enough to run in production.Many of these problems don't have established solutions yet. You'll read new research, test ideas, build evals, and ship the approaches that hold up.

What you'll work on

Agents and tool use: build multi-step workflows that turn a marketer's goal into concrete actions.

Generation: improve the quality of ads, emails, landing pages, and other marketing content produced by the system.

Evals: define what good looks like, measure it, and use the results to improve the system.

Context and memory: decide what a model needs to know at each step, what should be remembered, and what should be retrieved on demand.

Learning from outcomes: use marketer edits, feedback, and campaign results to improve future decisions.

Exploration: decide when the system should try something new and when it should stick with what is working.

Model systems: choose, route, and, where useful, fine-tune models while balancing quality, latency, and cost.

What we're looking for

• Strong ML fundamentals and experience building with modern LLMs beyond a prototype.
• Solid software engineering skills. Production reliability matters more here than a clever demo.
• Comfort taking an idea from a paper to an experiment to production code.
• Good judgment about which new techniques are worth the effort.

Relevant experience may include:

• Agents, tool use, or long-horizon reasoning.
• LLM evals and LLM-as-judge systems.
• Generative content systems.
• RAG, context engineering, or memory.
• Fine-tuning, post-training, or distillation.
• Contextual bandits, reinforcement learning, or online learning.
• Recommendation or explore/exploit systems.
• Model routing and inference optimization.

A background in computer science, machine learning, statistics, or a related quantitative field helps. An MS or PhD is useful but not required. You don't need any marketing or advertising experience.

Why this is interesting

Most LLM applications end when the model returns an answer. Different keeps going: the system needs to decide what to do, take action, see what happened, and use that result the next time it makes a decision.

In this role, you can test ideas against real marketing outcomes. The hard part is building a system that keeps working as the models, tools, and inputs change.

Logistics

San Francisco Bay Area 

Full-time 

5 days/week in office