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
Between a signed contract and a customer acting on our numbers sits the deployment. Enterprise procurement data is messy: the same material named four ways across plants, supplier records that disagree with the contracts, fields nobody can explain. You find the person who knows why, decide what maps where, wire the pipelines and get the first standardised outputs in front of category managers. Then you build on top. The work is funded as part of the platform and never billed as services, and the test for every task is whether it creates new value. You'll define how we deploy, and write the playbook the people after you follow.
What you'll own
- Schema discovery and mapping. Find out how their ERP and procurement systems hold the data, map entities to our model with documented field-by-field correspondences, and know what's missing as well as what mapped.
- Pipelines. Wire the extraction that keeps the model current, across integration tiers from flat file to read/write.
- First value live. Outputs category managers act on, tied to value they can point to.
- Building on the platform. MVPs, skills and account-specific views, agent-first. You decide what to build next from what you hear in the room.
- Bootcamps. Live co-build days on a prospect's own data, alongside our domain deployment lead.
- Productising what repeats. Any step you do twice becomes a template. Patterns worth keeping go to the product team, so each deployment makes the next one faster.
What we're looking for
- You've shipped production systems inside enterprise customers, recently. Embedded in their environment, with evidence on their side: a system in their stack, people who used it, a reference who'd say so. We can teach our platform; we can't teach the scar tissue.
- SAP-, Coupa- or Ariba-class systems. You've lived inside them and come out with working software, and the ugliest dataset you made sense of is a story you enjoy telling.
- Builder velocity. A strong generalist engineer (we lean Python) who gets an MVP in front of a user in weeks and makes the product calls yourself.
- Productising instinct. You know the difference between building on a platform and reaching into its core, even when the shortcut is tempting.
- Presence with the customer. You can run discovery and walk a procurement lead through what you built without a translator.
Nice signals: able to cover US customers as they come online.
Stack: Python, FastAPI, Postgres, Dagster, AWS; LangGraph, MCP, Pydantic AI. A signal, never a requirement.
The take-home: map a messy ERP-style extract to a canonical model, document the correspondences and tell us what's missing.
Forward Deployed Engineer, Product/Engineering employer: Atmospheric AI Limited
Atmospheric AI Limited fosters a collaborative work culture that emphasizes sustainability and innovation. Employees are encouraged to engage in impactful projects that contribute to the understanding of material sourcing and emissions. The company provides opportunities for professional growth and development in a dynamic environment.
