002 Hey there, I'm Oshen \o
My background is in search, information retrieval, and applied AI research. Over the past year, I led research & evaluations around how people use language models as search systems and how those models retrieve information, use sources and context, generate answers, and behave differently from traditional search engines.
The majority of my work has began with an unclear question about model behavior where I had to figure out how to turn that question into something I could observe, measure, and test.
However, the more time I spent evaluating model outputs, the more interested I became in the questions those outputs could not answer: what was happening inside the model, why certain behaviors emerged, and what those behaviors might mean for people and society as systems became more capable.
Outside of work and existential anxiety around super-intelligence, you can find me watching Jeopardy, going deep on a new hobby, tinkering with side projects, or reading.
The thread across my work is a curiosity about the systems that sit between people and information. In search, that meant understanding why certain results appeared, which signals shaped them, and how those systems influenced what people found and trusted. As my work moved into applied AI research, those questions expanded to include how language models select information, respond to context, and shape the way people understand and act on the models response.
I tend to work backward from an outcome, especially when the behavior is visible but the reasons behind it are not. I like taking an unclear question, finding what can actually be observed or measured, and building the research and evaluations needed to make finding an “answer” a little less mysterious.
That same thread is now leading me into AI safety, where understanding model behavior matters because it may reveal warning signs, expose weaknesses in safeguards, and give society more time to prepare for emerging risks. The context and stakes have changed, but my instinct remains the same: look closely at what the system is doing, question what the output actually means, and keep pulling the thread until the larger picture becomes clearer.
Areas of Expertise
I study how LLMs operate and behave as search systems, including how they interpret questions, retrieve context, generate responses, and influence how users search for, find, and interact with information.
I build reusable data pipelines, datasets, and predictive models that turn search, model, and product behavior into measurable signals.
I design research studies, evaluations, benchmarks, and scoring systems that make model behavior easier to test, compare, and improve.
I translate research findings into product direction, internal tools, dashboards, feature concepts, and clear guidance for technical and non-technical teams.
If something here resonated, I’d love to hear from you. I’m always open to talking about AI search, research collaborations, advisory work, or interesting problems that don’t fit neatly into a category.