Foundational Model Research Data Scientist

Sapience AI is a collective intelligence platform for professional communities that turns expertise scattered across CRMs, AMS platforms, and knowledge bases into searchable, actionable knowledge. The Foundational Model Research Data Scientist will design experiments, adapt and evaluate foundational models, improve grounding and trust, and translate research advances into dependable capabilities for the COGENT architecture and MINERVA platform.

Responsibilities:

• Design and run experiments on foundational models relevant to collective intelligence
• Investigate how models understand language, ground answers, and reason over knowledge
• Turn open questions into experiments with clear hypotheses and honest results
• Adapt foundational models to the language and needs of professional communities, including fine-tuning and alignment where it helps
• Improve grounding and reduce confident errors in domain settings
• Balance capability against cost, latency, and the constraints of production
• Build rigorous evaluation for what matters here: accuracy, groundedness, safety, and trust
• Design evaluations that reflect real community needs, not just public benchmarks
• Keep the organization honest about what a model can and cannot do
• Partner with data engineering on the datasets that training and evaluation depend on
• Handle data thoughtfully, including quality, bias, and protection of sensitive community knowledge
• Build the evidence base that makes model claims defensible
• Feed model advances into the neuro-symbolic COGENT architecture, and study how neural and symbolic methods work together
• Help decide where a foundational model belongs and where structure should carry the load
• Turn research into behavior the platform can rely on
• Track the fast-moving foundational model field and separate real progress from hype
• Bring in advances that matter and set aside those that do not
• Share knowledge so the whole organization stays current
• Study and reduce the failure modes that erode trust, including hallucination and bias
• Build toward models whose answers members can trust and trace
• Treat safety and trust as part of the research, not a later concern