Sr. Applied AI/ML Scientist

Description

NewsWhip by Sprout Social is looking for a Senior Applied AI/ML Scientist to join its AI, Data and Intelligence Business Unit.

Why join NewsWhip by Sprout Social's Data Science and AI team?

Most LLM applications are wrappers around a chat box. Ours aren't.

NewsWhip processes the world's news in real time - millions of articles, posts, and signals every day - and turns that firehose into predictive intelligence that journalists, PR leaders, and global brands rely on to make decisions before the news breaks. The interesting problems live everywhere in that pipeline: ambient agents that monitor and enrich content as it flows in, retrieval systems that have to be fast and correct on a corpus that changes by the minute, evaluation frameworks for a domain where hallucination has real-world consequences, and reporting workflows where LLM-generated insights need to stand up to scrutiny by professional analysts.

You'll work at the intersection of product, data, and engineering with the autonomy to make architectural decisions and the support of a company that takes AI quality seriously. If you want to build LLM systems that do more than answer questions - systems that reason, retrieve, enrich, and act across one of the most demanding content domains there is — this is the role.

What you'll do

Build and Ship LLM-Powered Features

• Architect and deliver production agentic workflows — both ambient (background agents that enrich, monitor, summarise, and surface insights across NewsWhip's data continuously) and interactive (user-facing tool-using agents that respond to journalist and PR analyst queries in real time).
• Own the design of LLM-powered content enrichments that feed directly into customer reporting, alerts, and intelligence briefings — turning raw news signals into structured, decision-ready outputs.
• Lead the technical direction for tool-augmented LLM systems, including MCP-compatible services, function-calling patterns, and multi-step reasoning workflows.
• Make architecture-level decisions about when to use retrieval, when to use agents, when to fine-tune, and when a smaller model or classical NLP approach is the right answer.
• Define structured prompting standards, output schemas, and reusable patterns that the wider engineering team can build on.

Drive Quality, Evaluation, and Observability

• Set the bar for how NewsWhip evaluates LLM systems — offline benchmarks, online experimentation, regression detection, and human-in-the-loop review.
• Own guardrails for safety, hallucination reduction, factual grounding, and output consistency in a domain (news and media intelligence) where accuracy is non-negotiable.
• Build observability into every LLM feature: tracing, cost tracking, latency budgets, quality metrics, and drift monitoring.
• Make pragmatic trade-offs between speed, cost, and quality, and communicate those trade-offs clearly to stakeholders.