Data Scientist

Dreamhub is the first AI-native CRM purpose-built for B2B SaaS teams. Backed by the founders of MuleSoft, Datadog, and Datorama, we’re redefining how go-to-market teams operate leveraging machine learning and LLMs to automate workflows, surface deep insights, and eliminate manual data entry.

You’ll own production-grade LLM-driven automation that reads customer interactions (emails, call transcripts, meetings) + CRM context and auto-populates CRM fields and automatically run workflows. You’ll ship end-to-end: spec → data readiness → workflow design → evaluation → production integration → continuous improvement. We’re a young startup with a small, highly experienced team (including people with past exits), and we’re looking for a Data Scientist who’s hungry for impact and wants real ownership.

Why DreamHub?

• Join a highly experienced team that has built and sold companies, and knows how to execute from zero to real outcomes.
• Work on real, high-impact problems at the intersection of AI and revenue operations.
• Own meaningful parts of the ML + LLM stack end-to-end (not “just modeling” or “just prompts”).
• Ship quickly, iterate often, and see your work directly influence customers and the product.
• Be part of a small, product-driven team building the future of AI-powered sales.

Responsibilities:

Own CRM field automation end-to-end

• Take full ownership of field automation projects: from defining field semantics and edge cases to shipping reliable automation in production.
• Design and implement LLM workflows using our internal framework (from single prompts to multi-step flows with code/tooling) to achieve strong precision/recall.

Drive “data readiness” and smooth product integration

• Work closely with Product + Engineering to ensure the right data is captured, accessible, and structured for training, evaluation, and runtime inference.
• Ensure automation flows plug cleanly into our broader ingestion and data pipelines, so that at the LLM workflow entry point you have the full context you need (emails/transcripts + CRM entities + metadata).
• Collaborate on implementation details to make the LLM-flow integration reliable, observable, and easy to maintain.

Evaluation, quality, and iteration

• Build and maintain evaluation sets, metrics, and error taxonomies (precision/recall, coverage, consistency, failure modes).
• Run systematic experiments to improve quality: prompt iteration, few-shot strategies, schema-aware prompting, retrieval/re-ranking, and structured output validation.
• Establish monitoring and feedback loops to detect regressions, measure drift, and improve reliability over time.

Engineering-minded execution

• Write production-grade Python: clean modular code, reproducible experiments, solid logging, and pragmatic testing.
• Partner with backend/data infra to ship robust integrations under latency and reliability constraints.

What Success Looks Like (First 3-6 Months):

• Multiple field automation flows shipped and trusted by customers (measured quality, clear monitoring, low nonsense-rate).
• A repeatable process for launching new automated fields (spec → eval → deploy → iterate).
• Reduced manual CRM work for GTM teams through reliable auto-fill.

Requirements:

• 2-3 years hands-on experience as a Data Scientist / Applied ML / ML Engineer.
• Strong Python and practical engineering habits (clean code, debugging, Git, building maintainable pipelines).
• Experience working with real-world messy data, especially text and semi-structured sources.
• Strong written and verbal communication (this role includes prompt writing, field spec clarity, and cross-functional alignment).
• High ownership: you can turn ambiguity into a plan, then deliver and improve.

Nice-to-haves (Big plus, not mandatory)

• Experience with LLMs / prompt engineering, including evaluation discipline.
• Familiarity with RAG, tool-calling/agents, text-to-SQL, or production LLM systems.
• Experience with evaluation frameworks, A/B testing, and monitoring quality in production.
• Relevant degree (CS/EE/Math/Physics/Stats) — not required if skills/experience are strong.