Lead Applied Scientist - AI Search & Brand Intelligence

SE Ranking Limited; Polska; On-site; PLN 350,000 - 550,000; Full time

Job summary

SE Ranking, an all-in-one SEO and digital marketing platform, is seeking a Lead Applied Scientist to reverse-engineer how AI search and generative platforms describe brands and to build a measurement layer proving AI visibility's value to businesses.

You'll run experimental pipelines, design sampling across topics and regions, and mentor a small ML team while translating insights into product features and industry-first research.

Qualifications

• 5-6+ years in Data Science or Applied Science with shipped decisions.
• Expertise in ranking and information retrieval (NDCG, precision@K).
• Strong GenAI/LLM experience using ML APIs and open-weight models.
• Rigorous evaluation: leakage detection, time-based validation, business metrics.
• Ability to lead a small team and set technical direction.

Responsibilities

• Design and run experiments to test hypotheses.
• Define features to improve AI visibility for brands.
• Develop sampling methods for AI responses across topics/regions.
• Establish evaluation framework for LLM outputs.
• Mentor ML Engineers and contribute to hiring.
• Turn findings into product features and research.

Skills

Data science

Ranking & IR

GenAI/LLM

Model evaluation

Technical leadership

English proficiency

Tools

ClickHouse

LightGBM

CatBoost

faiss

LLM APIs

Job description

SE Ranking, an all-in-one SEO and digital marketing platform, is looking for a Lead Applied Scientist to reverse-engineer how AI search and generative platforms (ChatGPT, Gemini, Perplexity, AI Overviews) recommend and describe brands - and to build the measurement layer that proves what AI visibility is actually worth to a business.

Why this role is different

• You get proprietary data most researchers can only dream about: SE Ranking's SERP, backlink, content, audit, GA and GSC datasets, plus Planable's social media data - combined with systematically sampled AI-generated responses
• Your output is not only product features. It's the GEO best practice the industry will end up using - with the opportunity to represent that work at conferences and in company research, if that's something you'd enjoy
• The discipline is barely a few years old. Nobody is "the expert" yet - the first rigorous answers will come from someone with your access to data

How the role splits

• ~45% hands-on research and modelling - you design and run experiments yourself
• ~45% technical and project leadership - hypothesis prioritisation, decomposition, methodology review
• ~10% people management - directing a small team's day-to-day (1-2 people): setting technical direction, running 1-1s, unblocking people

What you'll do - first 90 days

• Partner with our SEO/GEO specialists and selected engineers to set up a repeatable experiment pipeline: hypothesis → data scope → design → run → insight → decision
• Run the first 3-4 experiments and, with Product, define candidate features that would give our customers a real AI-visibility practice
• Design sampling methods to systematically collect and analyse AI responses across topics, intents and regions - with explicit handling of variance, bias and representativeness
• Establish an evaluation framework for LLM outputs, prompts and model behaviour, so results are comparable over time

What you'll do - ongoing

• Apply LLMs via batch API (including open-weight models) as feature extractors and judges: entity/sentiment/positioning extraction, prompt evaluation on holdout sets, cost-aware batch inference, agentic/tool-using pipelines
• Own the team's model evaluation & validation standards: leakage detection, temporal/out-of-time validation, and metric definitions tied to business outcomes
• Develop predictive and ranking models - gradient boosting (LightGBM/CatBoost), learning-to-rank (NDCG, precision@K) - for brand visibility, probability of mention in AI answers, and traffic/impressions in AI search and SEO
• Build classification, clustering and representation-learning models (embeddings, approximate nearest neighbours) that map how AI systems perceive and position brands
• Work directly in ClickHouse (or a similar columnar OLAP) on SERP, backlink, content, audit, GA and GSC data at hundreds-of-millions-of-rows scale, on infrastructure shared with the product
• Source hypotheses from across the company and the market, generate your own from patterns in data, and maintain a prioritised backlog with a clear execution cycle
• Mentor the ML Engineer(s) you work with: set technical direction, review methodology, delegate meaningfully, and participate in hiring
• Turn findings into product features, published research, conference talks and our own GEO best practice - working with Product, Engineering, Marketing and Leadership

What you'll bring

• 5-6+ years in Data Science or Applied Science, with a track record of taking ambiguous research problems from hypothesis to a shipped decision
• Deep, hands-on expertise in ranking & information retrieval (learning to rank, NDCG/precision@K, LambdaRank/LambdaMART) - the core lens for how brands get surfaced by generative AI systems
• Strong, current GenAI/LLM expertise: using LLMs as an analysis tool - prompting and evaluating both commercial APIs and open-weight models (e.g. Llama, Mistral, Qwen - models we can run on our own infrastructure), LLM-as-judge or agent-evaluation experience, and an understanding of how sampling, context and model behaviour affect what a model "says" about a brand
• Rigorous model evaluation and validation: leakage detection, temporal/out-of-time validation, and choosing metrics tied to the business outcome
• Comfortable directing a small team's day-to-day (1-2 people): setting technical direction, running 1-1s, unblocking people - this doesn't require management brilliance, just the willingness and basic ability to do it well
• Solid English (B2+) - able to explain a complex result to a marketer and a sceptical engineer in the same meeting

Nice to have

• Gradient boosting (LightGBM/CatBoost) for tabular modelling - we already have models built on this and need someone who can keep improving them
• Embeddings & representation learning (sentence-transformers, faiss) for semantic features and clustering
• Comfortable analysing large-scale data directly (e.g. ClickHouse) without relying on a dedicated engineer
• Genuine interest in the broader SEO/search domain - we don't expect domain expertise on day one, just curiosity and speed
• Familiarity with AI-search measurement concepts (AI Visibility, Share of Prompt, Share of Model)
• MLflow (or similar experiment tracking) experience
• Publications or talks (SIGIR, ECIR, KDD, ACL, EMNLP - or brightonSEO and industry research)

Mindset

• Growth mindset: you experiment without waiting for approval, adjust based on results, share what you learned, and turn failures into team learning
• Bias to action: you build the execution cycle first and refine the hypothesis backlog as you go, rather than spending a quarter defining the perfect hypothesis list
• Product-oriented: you think past the model, toward impact and actionability
• Critical thinking and the ability to disagree constructively
• Comfortable in ambiguous, cutting-edge problem spaces - and you escalates fast when blocked on data, rather than waiting

What we give you to succeed

• Direct, ticket-free access to SE Ranking's datasets - SERP, backlinks, content, audits, GA/GSC - plus Planable's social media data
• Dedicated data/analytics engineering support, so pipelines are not a solo project
• A dedicated LLM API budget for systematic response sampling (OpenAI, Google, Anthropic, Perplexity)
• An opportunity to publish: conference talks, research, and our own best practice
• Authority to launch experiments without asking for permission first

How we'll know it's working

• Cycle time from hypothesis to documented insight
• Actionability of the resulting insights - did they enter the product roadmap or the GEO best practice?
• Self-sufficiency in gathering and analysing data
• Number of experiments/ hypotheses run, and how many held up in practice and integrated into the product