HPC Engineer

EduRankAI; Kolkata District; In loco; INR 1.800.000 - 3.200.000; Tempo pieno

Descrizione del lavoro

EduRankAI in India (West Bengal, Kolkata) seeks an Engineer/Scientist to build scalable research software, simulate at large scales, and develop AI models for scientific problems with credible baselines. You will ensure reproducibility, data provenance, and robust testing across divisions.

The role requires strong Python/C++, Linux proficiency, and HPC tooling, with a focus on end-to-end delivery and transparent reporting of results.

Competenze

• Fluent in Python and able to write readable, testable code.
• Data structures and algorithms to justify choices.
• Comfortable on Linux and Git as a working tool.

Mansioni

• Build research software that others depend on and can be handed over.
• Build and maintain simulation platforms that scale to scientific needs.
• Optimize workloads with measured before-and-after results.
• Develop AI models for scientific problems with credible baselines.
• Create data pipelines with provenance to trace results.
• Maintain reproducibility: seeds, environments, and deterministic results.
• Ensure code quality and testing across repositories.
• Own end-to-end problem framing, execution, and reporting.

Conoscenze

Python

Algorithms

Linux

Git

Testing

Numerical methods

Baseline reporting

MPI

OpenMP

HPC scheduling

C++

Julia

Bash

Strumenti

MPI

OpenMP

HPC scheduling

Python

C++

Julia

Bash

Descrizione del lavoro

Making the department's workloads run at scale, with measured scaling curves rather than claims.

Department of Extreme-Scale, Nano & Fundamental Engineering, Division D08: Computer Science, AI & Scientific Computing. This division exists to serve all the others. It builds the research software, makes simulations run at the sizes the science needs, develops AI models for scientific problems with honest baselines, and holds the line on reproducibility and code quality across the whole department. This position: Making the department's workloads run at scale, with measured scaling curves rather than claims. Scale range for this work: 10^-100 m to 10^100 m. Research classification: Applied engineering. Engineering to a specification, on established science, for something that has to work. Evidential standing: Established science. Rests on settled physics and engineering. The work applies it; it is not what is under test here. A model that fits is not a model that predicts. Work from this division reports a held-out result and a competent classical or numerical baseline alongside every machine-learning claim, and states the domain a surrogate is valid over. A digital twin is a model of a system, not the system, and is described that way. Scale bands are a research classification, not a claim of experimental reach. Most of this span cannot be probed by any apparatus that exists: nothing below about 10^-19 m has been measured directly, and anything at 10^26 m or beyond is inferred from observation rather than engineered. Every opportunity states the kind of work it actually is. Level 3 (Engineer and Scientist). A whole problem end to end: framing it, doing the work, and reporting the result with its limits.

Responsibilities

• Build research software that other divisions depend on, to a standard that survives being handed over.
• Build and maintain simulation platforms, and make them run at the size the science actually needs.
• Optimise scientific workloads with a measured before-and-after, never an asserted one.
• Develop AI models for scientific problems, always reported against a competent baseline.
• Build digital twins, and state plainly what each one does and does not represent.
• Create data pipelines with provenance, so a number can be traced to the run that produced it.
• Hold the reproducibility standard: seeds, environments, and a result that comes back the same.
• Maintain code quality and testing across the department, including in other divisions' repositories.
• Own a whole problem end to end: framing it, doing the work, and reporting the result with its limits.
• Review a colleague's work when asked, and say plainly when a result is not supported by its method.

Skills

• Fluent in Python, and able to write code somebody else can read, test and extend.
• Data structures and algorithms to the level of choosing correctly and justifying the choice.
• Comfortable on Linux, and with Git as a working tool rather than a save button.
• Has written tests for numerical code and can explain what a tolerance in a test means.
• Numerical methods: can state the error behaviour of a scheme they have implemented.
• Reports a baseline alongside every model result.
• MPI
• OpenMP
• HPC scheduling
• Python
• C++
• Julia
• Bash