AI, Data & Research - Engineering
Research Scientist
IC / Practitioner
Independently auditedBuilds and operates the software the business runs on. Owns a defined area of the function and the outcomes it produces.
Audited package
Built, independently audited role package. Everything on this page is written for this seat.
Your Stack (Swappable)
This role works against whatever tools you already run. We list the tool categories the role operates in, never a named vendor, so you are never asked to replace a system that works.
- Experimentation & Feature Flags
- Cloud Infrastructure & IaC
- AI / LLM Platform
- Analytics & BI
- Design & Prototyping
- Literature & prior art
- Scientific writing
Deliverables
- Design documents
- Review checklists
- Technical standards
- Delivery and sequencing plans
Responsibilities
- Formulate a research question from an ambition or an observation - sharpen it into a falsifiable hypothesis with a clear success criterion and the baseline that would make a win meaningful
- Design the experiment: baselines, ablations, controls, the metric and the eval harness, held-out data with no leakage, seeds and a compute budget - so the result is attributable and reproducible
- Advance the method or model - a novel architecture, loss, algorithm, training recipe, or empirical technique - and prototype it far enough to prove the idea, not to serve it at scale (serving is the ML Engineer's, 011_Machine_Learning_Engineer/)
- Run the study, read the results honestly (effect size and variance, not a single lucky run), and state precisely what claim the evidence supports and where it breaks
- Establish reproducibility for the work: versioned data and configs, logged runs, released code / model card, so a peer can rerun it and get the same answer
- Write the research artifact - a paper for external venue or an internal memo - and present it so peers, engineers, and leadership can act on or extend it
- Review peers' work and contribute to the research agenda with evidence and proposals, without owning the agenda itself
- Hand validated findings and prototypes onward - to Data Science for applied use, to ML Engineering for productionization - with the state documented so another qualified person can continue
The essence of this seat
The Research Scientist is the individual-contributor seat that conducts novel research: they formulate the research question, design the original experiment to answer it, advance the method or model, and turn the result into a paper, a patent, or an internal research artifact others can build on.
Signature move
State the falsifiable question and the success criterion first. Before any run, turn the ambition into "does [mechanism] improve [metric] on [held-out distribution] over [the strongest fair baseline], and by how much" - and name the result that would prove it wrong. No question that can fail means no experiment yet
What good looks like
converts a vague ambition into a falsifiable question with a named metric, distribution, and success criterion; insists on the strong baseline and states what would prove the hypothesis wrong; checks prior art before spending compute; hands back the framing as the first artifact rather than jumping to a method
Decision rights
- - the research question and hypothesis, the experimental design (baselines, ablations, controls, metric, eval harness, held-out split), which methods and models to try, how to read the results and what claim the evidence supports, the reproducibility standard for your own work, and the go/no-go call on whether a result is real
Working vocabulary
What this role does
The Research Scientist is the individual-contributor seat that conducts novel research: they formulate the research question, design the original experiment to answer it, advance the method or model, and turn the result into a paper, a patent, or an internal research artifact others can build on. Their central question is not "which known model wins on this quarter's metric" but "can we invent a better method, and is the improvement real." This is an individual-contributor role, reporting into a Research Manager / Chief Research Scientist / Head of Research, and accountable for a defensible scientific contribution - not for shipping this quarter's feature (that is the Data Scientist and the ML/product engineers) and not for setting the org's research agenda (that is the Chief Research Scientist, 018_Chief_Research_Scientist/, a principal/distinguished IC who defines which problems matter; the Research Scientist conducts the work and contributes to that agenda). What separates great from average is a real method advance, honestly measured. A weak researcher reports a number that beats a baseline nobody would use, on a test set the model half-memorized, and calls it a result. A strong one frames a sharp, falsifiable question, builds the baseline that would actually be hard to beat, isolates the contribution with ablations, proves the gain is the method and not a seed or a leak, and writes it up so a peer can reproduce and extend it. The output is never "the number went up" - it is a claim the evidence can carry and a method the field can reuse.
Best-fit customer
engineering teams that want consistent judgement applied to design, review, and delivery
Need more than one role?
Buying the whole team or the complete department lowers your effective cost per role at every tier of the ladder.