Trading & Pricing Scientist

Everything physical is made from raw materials: polymers, chemicals, metals and minerals, a $16T economy that software has barely modelled. Atmospheric AI computes what those materials should cost, what carbon they carry and where their supply is exposed, from the physics of how they're made, so manufacturers can decide and negotiate from ground truth.

The problem

Benchmarks are public. The price a converter in north-west Europe paid a Middle East supplier for a film grade isn't. Procurement teams need that number, with a range they can defend, before they sit down with the supplier. You'll build the models that estimate it.

What you'll own

• Transaction-price inference. Estimate what a buyer at one facility likely paid a supplier at another, given the grade, the geography, the market structure and the likely contract.
• Pricing mechanics. Represent PRA benchmarks, formula and cost-plus contracts, grade differentials, regional premia, netbacks and captive transfers.
• Should-pay ranges. Turn estimates into explainable ranges customers take into supplier negotiations, with the evidence and uncertainty on show.
• Price signals. Source, version and maintain the benchmark, feedstock and utility prices used across the platform.
• Validation. Test estimates against withheld prices, report error and bias, and say plainly where the model can't support a claim.
• Production. Ship the models as tested Python modules, and shape how we add forecasting as the pricing layer matures.

What we're looking for

• You know how opaque physical commodities are priced and contracted, beyond how their futures trade. We can teach our stack; we can't teach years in these markets.
• Quantitative modelling under sparse, noisy data, and the judgement to know when more complexity stops paying.
• Production-quality Python. Interfaces, tests, review, reproducibility.
• Commercial sense. You can explain to a procurement lead why the benchmark isn't the price they pay, and give a useful range when the market won't support a precise one.

Nice signals: more than one commodity complex; Argus, ICIS, Platts, Fastmarkets or CRU data; should-cost or negotiation work; price forecasting; a quantitative Master's or PhD. Experience only in liquid financial instruments is unlikely to be enough.

Stack: Python, FastAPI, Postgres, Dagster, AWS. A signal, never a requirement.

The take-home: estimate the fair delivered price of a named polymer grade between two regions, with a range, the evidence behind it, how you'd test it and what would change it.