AI Product Manager

Carrum Health•,

2d•Hybrid

About The Position

At Carrum, we are transforming how we pay for, deliver and experience healthcare. We are looking for an Applied AI Product Manager to help us lead the development and execution of AI-powered healthcare products. The Applied AI Product Manager will play a critical role in our success, by bridging the gap between technical teams, clinicians, and business stakeholders, turning complex AI capabilities into actionable, user-centered solutions. The ideal candidate has hands-on experience in applied AI product development, an entrepreneurial spirit, and a proven ability to deliver products from concept to market. You will own key AI product features and their delivery and be crucial in translating clinical and business needs into practical AI solutions; helping us ensure that our technology truly improves our member experience and supports our care teams. We are building an Applied AI team, and are seeking a Product Builder who can turn raw model capabilities into intuitive user experiences. You are likely a former engineer or data scientist who realized you care more about what we build and why, rather than just how. You don't just write PRDs; you open the IDE, test the APIs, prompt the models, and look at the raw data before you ever open Jira. You believe that in the era of AI, the gap between "Idea" and "Prototype" should be measured in hours, not weeks. You’ve hustled, you’ve created, you’ve implemented, and you love the intensity of a startup in high-growth mode.

Requirements

4–7 years of product experience, with at least 2+ years specifically building and shipping AI/ML products.
Strong preference for a CS background or previous experience as a software engineer/data scientist.
Working knowledge of the modern AI stack (LLMs, RAG architectures, vector databases, fine-tuning vs. context injection).
Ability to read technical documentation and understand API capabilities/limitations independently.
Experience navigating the complexity of healthcare data. You understand that "patient safety" is the ultimate constraint.
You operate comfortably at every altitude. You can present a quarterly roadmap to executives, model the unit economics of AI features, and dive into the weeds to debug a prompt strategy—always balancing a "magical user experience" against commercial viability and cost
You are obsessed with closing the gap between model performance and real-world adoption, behavior change, and user / business impact.
You are comfortable working in a fast-paced dynamic environment and keeping many balls in the air, and are resourceful and willing to find creative ways to make a big impact quickly.

Nice To Haves

Proven Builder: You have shipped complex AI features well beyond simple API wrappers, with a proven track record of leading solutions that involve RAG (Retrieval-Augmented Generation), multi-step agentic workflows, and advanced LLM architectures
A Tinkerer at Heart: You have a GitHub profile, a folder of side projects, or a history of building your own tools. You understand the "texture" of AI - you know why a model is hallucinating because you’ve wrestled with it yourself.
Data-Native: You don't wait for a dashboard. You are comfortable running SQL queries, inspecting JSON outputs, or looking at raw logs to understand user behavior and model failure modes.
Pragmatic over Hype: You know the difference between a cool demo and a production-ready feature. You obsess over latency, cost, and reliability.

Responsibilities

Prototype to Spec: Instead of writing abstract requirements, you will build functional prototypes using tools like Streamlit, LangChain, or Python notebooks to validate feasibility. You verify prompt strategies in playgrounds (OpenAI, Anthropic, Vertex AI) before engineering writes a single line of production code.
Bridge the Technical Gap: Translate "model constraints" (context windows, inference costs, probabilistic outputs) into concrete product mechanics. You will sit with engineers to debug edge cases, not just report them.
Own the "Applied" in Applied AI: Drive the transition from "model performance" (F1 scores) to "product performance" (user success rates). You will define what "good enough" looks like for production.
Wrangle Complex Data: Work hands-on with messy, real-world healthcare datasets (claims, EHR, patient-reported data). You view data quality not as someone else's problem, but as a core product asset.
Design for Trust, Speed & Explainability: Collaborate closely with Compliance, Design, and Engineering to build UIs that help users understand AI outputs, handle probabilistic edge cases gracefully, and foster user trust in the system. You also understand the critical tradeoffs between model accuracy and speed to insight, and design with this in mind.
Ship Responsibly: Design the guardrails, fallback mechanisms, and human-in-the-loop workflows that make AI safe and compliant in a healthcare setting.