Oxiedo is building the architecture that lets a regulated institution prove what its models did. The roles below are the ones that decide how fast that reaches the market — commercial leadership, the experimental programme, and two specialist engagements that gate specific applications.
Each one states what it owns and what it requires. The commitment shape is on every listing, because a research partnership and a full-time equity role are different propositions and should not be described as though they are the same.
- Stage
- Pre-seed, commercially open
- How we work
- Direct. No recruiters, no screening layer, no take-home
- Location
- Remote. The founder is based in Dhaka
- Reply time
- A few days, usually less. Every message read by a person
- 01
Dataset partner
Priority engagement
We are looking for an institution holding data it cannot currently vouch for.
Two situations qualify: a labelled corpus suspected of being dirty, or a training run that degraded in a way nobody could account for afterwards. Those are the two conversations we most want to have.
We run the diagnostic and the institution receives the report. The data never leaves the customer environment, because the architecture has no outbound route. That is a property of how it is built rather than an undertaking anyone has to take on trust. Findings are published jointly, or not at all, at the institution's discretion.
What we require: authority to approve a research engagement, and a corpus with a real obligation attached to it: clinical, financial, or held under a deletion clause.
Apply for this - 02
Commercial lead
Commercial leadership across the regulated markets this sells into.
This role belongs to someone who has stood on the buying side of the table in one regulated domain: inside a hospital’s AI group, a bank’s model risk function, a biotech’s data organisation, or coordinating a multi-site research consortium. Knowing how that budget is released, and who has to sign, is the qualification.
The commercial side belongs to this role entirely: first engagements, partnership terms, and the relationships that decide which market opens first.
What we require: a network in one of the five markets that cannot be built from the outside, and enough technical fluency to hold a CTO’s questions without needing to be the architect.
Apply for this - 03
Research engineer
Ownership of the experimental programme that extends the architecture.
Two questions sit at the centre of it. Does the per-component health signal lead the global loss curve, and by how much? And does inter-component disagreement separate mislabelled data from data that is merely hard?
The second decides how far the product reaches, which is why it is scheduled before the build rather than after it. The role designs the experiments capable of falsifying the thesis, and co-authors the result either way.
What we require: published experimental work, fluency in PyTorch at the architecture level rather than the training-loop level, and the discipline to run a study that could return a negative.
Apply for this - 04
Regulatory consultant
One judgement is needed: whether the artefact we produce is the artefact a reviewer expects.
The work is bounded: read a specification and say whether the record matches what a reviewer would accept, or whether it is something adjacent they would not recognise. Either answer is worth the fee.
It gates the largest application in the product, so it is scheduled early rather than late. A wrong format is cheaper to find now than after a submission is built on it.
What we require: a filed change-control plan for a machine learning system under the FDA’s PCCP framework, the EU AI Act, or an equivalent regime. Filed one, not advised on one.
Apply for this - 05
Data-protection counsel
One question to settle: what may and may not be certified when data is removed from a trained model.
The question is specific. If a bounded structure trained on a named sample set is deleted, and a measured bound on the residual influence of that data can be stated — what may we certify, what may our customer certify onward, and does any of it constitute erasure under Article 17?
A certificate whose wording has not been tested creates liability rather than removing it, which is why this is answered before anything is issued.
What we require: current practice in AI and data rights, and willingness to put a position in writing.
Apply for this
No recruiters. No tracking system. No take-home.
Name the engagement and the relevant work behind it. A paragraph is enough. A CV is welcome if one is to hand, and not required.
If it is not a fit we say so directly rather than going quiet. If it is, the next conversation is with the people who built the architecture, with no screening round in between.
One form, and it reaches the people who built the architecture.
No tracking system, no automated screen, no acknowledgement email written by somebody else. A poor fit gets a direct answer rather than silence to interpret.