Job description

Nobody calls a logistics coordinator to say things went well. They call because a $100,000 shipment of clinical trial medication has been sitting in customs for three days and nobody can explain why. Because a biologic therapy arrived outside its temperature window and the treatment has to start over. Because the patient is waiting and nobody in the supply chain can give a straight answer about the shipment’s location. This is the problem we’re solving. We’re a year in, tracking over 500,000 shipments (including for Fortune 100 customers), and we’ve already cut manual ops effort in half. General Catalyst, Eclipse Ventures, and Virtue led the seed. The angels are all former or current operators who've spent careers tracking down shipments themselves, such as Head of Logistics at Bristol Myers Squibb, CMO of Cardinal Health, President of Novo Nordisk US, President of UPS Air, and CEO of Uber Freight. They know the market is as large as the problem is broken. What we’ve built works, and the most interesting problems are still ahead of us. - Olivier, Co-founder & CEO THE ROLE Founding data scientist / ML Engineer. The platform sees more about a shipment than anyone else in the supply chain does. Your job is to turn that into prediction: models that flag the customs hold, the temperature excursion, and the silent carrier before the issue reaches a patient. This is the Predict in our See/Predict/Act framework, and it is the part customers cannot get anywhere else. You'll own it end to end: the models, the data and feature infrastructure they run on, and the experiments that prove they work. You'll define how we measure a model, not just build one. Every person at Zoomlogi talks directly to customers, and you will too. The ops manager who tells you which exceptions actually hurt is the best feature-selection input you'll ever get. WHAT YOU'LL BUILD - The risk and prediction models. The platform ingests a continuous stream from carriers, forwarders, IoT sensors, weather, and flight data, stitched into a unified picture of every active shipment. You'll refine the models that scan that picture for risk and flag issues before they cascade, across hundreds of carriers, lanes, and sensor types. The hard part is the long tail and the asymmetry: a missed signal is measured in patient outcomes, and a noisy one trains operators to ignore you. Getting precision and recall right on real shipments is the work. - Dynamic re-routing optimization. Detecting a problem is half the job. The other half is deciding what to do about it: which alternate routing, carrier, or mode actually saves the shipment, and whether the fix is worth its cost and risk. You'll build the optimization that turns a flagged exception into the best recoverable plan, balancing transit time, temperature exposure, cost, and the hard constraints a healthcare payload carries. This is the Act in See/Predict/Act, and it is where a model stops being an alert and starts being a decision. - The data and feature infrastructure. A model is only as good as what feeds it. You'll own the pipelines and feature layer that turn 50+ messy external sources into model-ready signal at low latency. Some sources claim a shipment is delivered while the GPS shows it still in transit. Some go silent without warning. Building features that hold up against that, and that the next model can reuse, is yours. - The experimentation system. How do we know a model is actually working on live shipments? You'll build the offline evaluation and online experimentation that answers that: backtesting against real outcomes, measuring lift, and catching drift before customers do. You'll define the metrics the rest of the team trusts. WHO THRIVES HERE - You've shipped a model to production and watched what it did there. You have a specific recent example of something you built, owned, and put in front of real decisions, and you care about what happened after it shipped. Notebook-to-nowhere work is not what this is. - You're comfortable with ambiguity at the problem level. The interesting problems here don't arrive as labeled datasets with a defined target. You hear something from a customer, decide what's worth predicting, find or build the signal, and ship it. If you want the problem fully specified before you start, this role will frustrate you. - You find the domain genuinely interesting. The compliance constraints, the quality and logistics tension, and the fact that your model's output moves medication to real patients should read as compelling design constraints, not obstacles. - You're honest about what you don't know. We debate hard, change our minds, and challenge each other, always assuming everyone in the room is trying to get it right. A model that looks good for the wrong reason is something you want to catch, not defend. - You don't think customer contact is a tax on your time. The ops people on those calls will tell you which errors actually cost something. That is signal you can't get from the data alone. This role is probably not right for you if you prefer to go deep on one research problem and be left alone, or if you measure success by offline metrics rather than shipped impact. The surface area is wide: detection models, feature infrastructure, experimentation, and agent evaluation. If switching between modeling and infrastructure drains you, this will too. WHAT WE'RE LOOKING FOR You have 3-9 years of applied data science or ML experience. You've put models into production, with the evaluation and monitoring that keeps them honest, not just trained them offline. You're strong in Python and the modern data and ML stack. You're comfortable owning data infrastructure: pipelines, feature engineering, and the messy work of making real-world data usable. You have real experimentation rigor and know the difference between a model that scores well and a model that works, and you can design the test that tells them apart. You understand data modeling and how your work connects to the systems around it. You're based in San Francisco or Chicago and excited about being in the office. THE TEAM Founded by a team of operators with deep experience in Logistics Tech & Healthcare (Ex-Uber Freight GM + Airspace CRO), joined by early engineers from the likes of Uber, Hippocratic AI, Assort Health and others. We’re here to solve a real-world problem at scale, by making every critical shipment visible, predictable, and on time, so potentially life-saving therapies reliably reach the people who need them. THE STACK Python, React/TypeScript, Kafka, AWS. The data layer runs on a continuous stream stitched across 50+ sources. We do a lot of low-fidelity prototyping, Google Slides included. LLM infrastructure and orchestration are increasingly central to how we build. THE INTERVIEW PROCESS We move quickly. - Intro call: role fit, motivation, what you've shipped - Technical screen: a realistic modeling and data problem close to the actual work - Onsite: we'll work through a detection or prediction problem together - References then Offer COMPENSATION & LOGISTICS - Base salary: $180K-$260K - Equity: Above market (4-year vesting, 1-year cliff) - Benefits: Best-in-class medical and dental (100% self, 50% dependents) - Location: San Francisco HQ (Mission) or Chicago (The Mart). Hybrid, 4 days/week in office. No remote application accepted

Requirements

Must-haves

What we’ve built works, and the most interesting problems are still ahead of us.
Olivier, Co-founder & CEO
You'll own it end to end: the models, the data and feature infrastructure they run on, and the experiments that prove they work. You'll define how we measure a model, not just build one.
Every person at Zoomlogi talks directly to customers, and you will too. The ops manager who tells you which exceptions actually hurt is the best feature-selection input you'll ever get.
You don't think customer contact is a tax on your time. The ops people on those calls will tell you which errors actually cost something. That is signal you can't get from the data alone.
We move quickly.
Intro call: role fit, motivation, what you've shipped
Technical screen: a realistic modeling and data problem close to the actual work
Onsite: we'll work through a detection or prediction problem together
References then Offer
Base salary: $180K-$260K
Equity: Above market (4-year vesting, 1-year cliff)
Benefits: Best-in-class medical and dental (100% self, 50% dependents)
Location: San Francisco HQ (Mission) or Chicago (The Mart). Hybrid, 4 days/week in office. No remote application accepted

Nice-to-haves

None flagged.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

mlgorestpythonaireacttypescriptawsllm

Platform: ashby
Effort: quick

Benefits

Benefits detected in the listing.

Dental