Senior Data Scientist in United

Python, SQL, and LLM APIs. Experience with an agent or LLM orchestration framework (LangChain, LangGraph, LlamaIndex, DSPy, or comparable) and with retrieval and embedding tooling.

You will own the modeling and intelligence layer of Rolodex. You’ll turn the data generated by our searches (consultant actions, candidate and client interactions, notes, transcripts, references, and outcomes) into signals that help our consultants make better decisions.

\n

\n

This is a hands-on role. You’ll build models, put them into production with our engineering team, measure whether they work, and use the data generated by subsequent searches to make them better.

\n

\n

What you will work on

\n \n

1. Take the existing models from static logic to self-improving systems. How we surface and prioritize candidates for a search should sharpen every time a consultant takes actions in Rolodex and searches progress. \n

2. Own the unstructured-text layer. Prompt engineering, named entity recognition, and extraction across call transcripts, consultant notes, and references. You will design systems and mine our data sources for new insights. \n

3. Build new models to power features like similar candidates, similar companies, and similar jobs, leveraging features from our proprietary data. \n

4. Close the learning loop. Feed the outcomes of every executed search back into the models so the engine compounds rather than persists. \n

5. Make every model explainable. Our consultants and clients need to understand why Rolodex surfaced, ranked, or recommended something, and be able to interrogate the evidence behind it. You’ll build models that expose the signals driving their outputs, where those signals came from, and where the model is confident or uncertain. \n

\n

\n

What you have done

\n

We’re looking for someone who has:

\n \n

• Built and shipped a production scoring or ranking system that real people used to make decisions. You can walk through why it was designed in the way it was, what you’d change now, and whether it beat what it replaced. \n

• Taken a model you built through engineering’s release process to production. You can describe the handoff: what you packaged, what broke, and what you changed so it would land. \n

• Turned messy unstructured text like transcripts, notes, and free-form prose into structured features or fields using LLMs. You can describe how you knew it was working: what you evaluated against, and what it changed downstream. \n

• Built models whose outputs needed to be understood and trusted by the people using them. You can describe how you made the model’s reasoning explainable, how you surfaced the evidence behind its outputs, and how that affected adoption or decision-making. \n

• Worked directly with non-technical business users on a problem that arrived undefined, and delivered something those users adopted. You can describe how you worked out what was needed and what changed in how they worked. \n

\n

\n

The Cole Group places CROs, CMOs, and go-to-market leaders at high-growth software companies like OpenAI, Cursor, Lovable, Clay, Replit, Vercel, Kalshi, among many others of the most recognizable names in AI and B2B technology.