A guide to hiring a Fortran developer
Fortran is one of the oldest programming languages, and it was made for numbers. It is still used for scientific and engineering programs such as weather and climate models, physics and chemistry simulations and structural analysis, mostly because the code already exists and works. A Fortran developer keeps that code running, makes it faster and easier to change, and connects it to newer tools. If your team depends on a program nobody fully understands, or a model that takes too long to run, ask for Fortran experience by name.
What the job covers day to day
New code is only part of the work. A Fortran developer reads existing routines, finds a bug in a calculation, adds a feature to a model, updates the build for a new compiler and checks that a change did not move the results. A lot of time goes to talking with scientists and engineers, because the person who knows the science and the person who knows the code are often different people. Clear notes and plain explanations matter as much as the code itself.
Old and new Fortran
Fortran code comes in very different styles. Older programs are written in fixed form, with strict column rules, shared data blocks, jumps between lines and variables that are typed by their first letter. Modern Fortran has modules, derived types, allocatable arrays, whole-array operations and clear interfaces. Most real projects hold both. A good developer can read the old style without fear and write the new style with care, and knows how to move from one to the other without breaking anything.
Why the numbers matter
In numerical code, a small change can alter a result. Adding numbers in a different order can change the last digits. A different compiler setting can change how a calculation is rounded. A careful developer knows this, saves reference outputs before they touch anything and agrees with you how close a new result must be to the old one. They also say when a difference is harmless and when it points to a real mistake.
Modernizing step by step
Few teams can stop work for a rewrite. The safer path is gradual. First add tests that capture today’s behavior. Then turn on stricter compiler checks, declare every variable, replace shared blocks with modules, split long routines and add clear interfaces. Each step is small and is checked against the reference results before the next one starts. The code stays usable the whole time, and the team learns it as it improves.
Speed and memory
Slow runs cost time and money, and they limit the questions a team can ask. A developer who works on speed starts by measuring. The cause is often how arrays are laid out in memory, a loop that does the same work many times, or a compiler option that was never turned on. Only after that do they think about running on more cores or more machines. Ask a candidate to tell you about a run they made faster, and what they measured before and after.
Parallel and GPU work
Large models are often split over many processors. OpenMP shares work across cores in one machine, MPI passes messages between machines in a cluster, and coarrays are part of the language itself. Some codes also move loops to a GPU with directives. Parallel code brings its own bugs, such as races and uneven load, so it should be added carefully and tested often. Our MPI developers focus on this side of the work, and a Fortran developer works well beside them.
Fortran with Python and C++
Many projects keep a Fortran core and wrap it in newer tools. Python is popular for setting up a run, reading results and drawing plots, and C and C++ are common for other parts of a system. Fortran has a standard way to talk to C, and tools exist to call Fortran from Python. The developer sets up these links and checks that sizes, types and array order match on both sides. See our Python developers and C++ developers for the other half of such projects.
Building and testing
A program that only one person can build is a risk. A Fortran developer tidies the build with Make or CMake, makes sure it works on more than one compiler and writes down the steps. Tests come in two kinds: unit tests for key routines, and regression tests that run whole cases and compare the output with agreed tolerances. Together they let a team change code with confidence and spot problems early.
When a Fortran developer is a good fit
- You rely on a scientific or engineering program written in Fortran.
- The person who wrote the code has left, and nobody is sure how it works.
- Runs take too long and you want them faster before buying more hardware.
- You want to connect an old model to a Python workflow or a web tool.
If you are starting a new project with no existing Fortran, another language may be the easier choice. A developer who says that out loud is giving you honest advice.
Interview questions worth asking
- How would you change a long routine without changing its results?
- Why can the same sum give slightly different answers on two machines?
- What do you check first when a run is slower than expected?
- How would you replace shared data blocks with modules in a safe way?
- How do you pass an array from Python to a Fortran routine?
Good answers come with a reason and a past project behind them. If someone knows the syntax but cannot say how they would prove a change was safe, keep looking.
Try a small paid job
Hand every finalist the same piece of your real work: a short routine with a known bug, say, or a request to add a test around one calculation. Pay them for it. Then compare the work, the questions each person raised and how plainly they explained the result.
Setting up for the long run
Settle the routine in week one: where the code lives, who reviews a change, how the tests are run and who signs off a change in results. Keep the code in version control and write a short note on how to build it and what each main module does. That note saves the next person many hours, and it keeps the project from depending on one expert.
Mistakes to avoid
- Changing code with no saved results to compare against.
- Rewriting everything at once when a gradual path would do.
- Tuning for one compiler and one machine only.
- Judging by the list of languages on a resume, with no real numerical project behind it.
Related help: our machine learning engineers can work with the data your models produce, and our software developers help build the tools around your code.















