About Curium

Curium is the enterprise platform for brands to win AI search. Founded by the Princeton team that invented and coined GEO (Generative Engine Optimization), we combine proprietary reinforcement-learning models with a turn-key content optimization platform. Our customers range from household-name consumer brands and a major internet media company to one of the largest US sportsbooks, fast-growing fintechs, and category leaders in energy, education, and real estate. Backed by General Catalyst, the ex-CTO of OpenAI, and the CEO of Intel.

The team

Curium was founded by the Princeton researchers who defined GEO. Co-founder Ameet Deshpande, an Olympiad medalist, went from IIT Madras to a PhD at Princeton. Co-founder Karthik Narasimhan is a Princeton professor whose work includes OpenAI's original GPT paper, ReAct, and SWE-bench, and co-founder Vishvak Murahari came up through Princeton and Google DeepMind. They're joined by engineers from Apple, MIT, and UC Berkeley. We're a small team in San Francisco where everyone writes code, talks to customers, and ships.

What we've built

Search is moving from ten blue links to one generated answer. Generative engines like ChatGPT, Gemini, Perplexity, and Google AI Overviews retrieve sources, synthesize them, and decide which brands get cited, recommended, and described, and how. That process is a black box. It's non-deterministic, and it shifts with every model update. SEO tools were built to rank pages against a known algorithm, and they have no model of how an LLM reasons about a brand.

We treat GEO as an optimization problem. Curium continuously probes AI engines across the queries that matter to a brand and measures visibility: whether the brand is cited, where it appears in the answer, and how it's described. On top of that, we built proprietary reinforcement learning algorithms that learn which changes to a brand's content, structure, and data actually move those answers, using measured visibility in the engines as the reward signal. The system proposes a change, we ship it, the engines respond, and the models improve with every deployment. This builds on our GEO research, published at KDD 2024.

About the role

You will build the product enterprise brands use to win AI search, end to end, from the database to the UI. This is a small team with high ownership: you take features from a customer conversation to production, and you are judged on whether customers use what you ship, not on how many tickets you close. You will ship directly with the founders and join customer calls, so you see exactly how your work lands.

What you'll do

• Ship features end to end. Own features across the backend and frontend, from data model to UI, and get them in front of customers fast.
• Build the integrations. Connect Curium to the platforms our customers run on, starting with enterprise CMSs, and make them reliable at scale.
• Build on top of LLMs. Develop the systems that adapt Curium's output to each brand's voice and guidelines, and make them hold up in production.
• Partner on ML and data. Work with our ML team to take high-impact ML and data projects from prototype to production.
• Stay close to customers. Join customer calls, see how the product is actually used, and let it change what you build next.

This is a startup. You will own your features, but you will also be pulled into whatever needs to get done. If you need a clearly scoped job description to be comfortable, this isn't the right fit.

Who you are

• 5 to 10 years of engineering experience.
• Full-stack. You have shipped across the backend and frontend, and you are not siloed to one layer of the stack.
• You move fast with a small team. You would rather ship and learn than plan and wait.
• Product sense. You care about what the user sees, not just what the system does.
• Customer-aware. You want to see how a customer uses the thing, and it changes what you build next.

Nice to have

• Startup experience. You have worked at an early-stage company and owned outcomes, not just tickets.
• You have shipped products or features on top of LLMs in production.