Applied Scientist

Company Description FalconFirst is a decision intelligence company building an AI-native decision layer for product teams. Instead of shipping features based on gut feel and noisy opinions, we help teams validate ideas inside real workflows before development - using prototype interactions, user feedback and experiment signals to produce decision-grade outputs. ⸻ Role Description If you want predictable hours and a slow pace, don't apply.If you want to build the "brain" of a new product category and ship weekly, this is for you. You'll help build the intelligence layer behind Falcon Terminal: systems that convert messy behavioral signals + unstructured feedback into Decision Cards; clear claims with evidence, uncertainty and recommended next tests. You'll work across inference, agent workflows, evaluation/guardrails and instrumentation foundations. This is not a "model in a notebook" role. You'll ship production systems, iterate weekly and influence core product direction. How we work: fast cycles, direct feedback, high ownership. Not a 9–5. ⸻ Responsibilities (high-level) • Turn prototype interactions + feedback into structured evidence and decision-ready insights • Build inference that quantifies uncertainty and improves as evidence accumulates • Design lightweight NLP pipelines (clustering/embeddings) and use LLMs pragmatically (with evaluation) • Build evaluation + guardrails so outputs are grounded, consistent and reliable • Collaborate with engineering on instrumentation/SDK primitives and workflow orchestration • Help define what "good" looks like (quality metrics, stability checks, regressions) ⸻ Qualifications Must-have • Strong fundamentals in statistics/inference (Bayesian + experimental design is a plus) • Strong Python (SQL is a plus; R/Stan/PyMC helpful) • Practical experience with NLP (embeddings, clustering, summarization) and an evaluation mindset • High agency: you move from ambiguity → shipped system quickly • Ability to communicate clearly under uncertainty (PM-friendly explanations) Nice-to-have • Fine-tuning/distillation or prompt+eval pipelines • Product analytics experience (funnels, cohorts, paths) • Reliability patterns for LLM/agent workflows (routing, fallbacks, caching, monitoring)