Raihan Abdul Vaheed

Architected a Python LLM Toolbox middleware library — extensible tool APIs, safety scopes, and prompt integration workflows for agent applications — leading a team of three.

I owned the CI/CD as well: GitHub Actions for linting, type checking, automated testing and release automation, plus dependency management and pre-commit standards. A good chunk of the role was sitting with prospective clients and turning what they described into a scope someone could actually build against, tracked as GitHub milestones.

Python, LLM Agent Design, Pydantic, GitHub Actions, CI/CD

Owned the data engineering side of a spiking neural network research project, using Apache PySpark to build distributed preprocessing pipelines over macaque motor cortex electrode recordings before they reached a model.

Co-led the team translating neural activity into muscle control commands, applying signal filtering and deep learning to model brain-controlled arm movement for prosthetics and BCI applications.

Apache PySpark, Python, PyTorch, Signal Processing

Built the application backbone that enabled on-device software development for the Axon-R brain-computer interface hardware, working across Android Studio (Java/Kotlin), Unity (C#) and Python.

Developed Unity applications for the platform as one of three engineers, and set up and maintained my own build and deployment pipelines alongside documentation, testing and ongoing system support.

Java, Kotlin, C#, Unity, Python, Android Studio

Co-designed an automated test suite in Playwright and TypeScript to validate the application data pipeline, and integrated it into CI so regressions were caught on every build. Two of us covered quality for the whole product.

The work was verification and validation — unit, module and static analysis — and I leaned on neuroimaging domain knowledge to improve coverage and debug the application. In that domain silent failure isn't an acceptable outcome, and it shaped how I've thought about test discipline everywhere since.

TypeScript, Playwright, CI/CD, V&V

I spent weeks unable to make my optimizations to a Python machine learning API fast enough. Rather than settle for what I had, I went back to my math professors to properly understand the concepts underneath the problem. The rewrite that came out of it ran up to four times faster through better parallelism.

What stayed with me was the diagnosis rather than the fix. Bottlenecks are almost never where you first assume they are, and you find them by measuring instead of guessing.

I also used Apache PySpark to simulate distributed workload scenarios against the platform, contributed to compiler development using concurrency techniques, and maintained my own build pipelines.

Python, Apache PySpark, Multiprocessing, Compilers, MindSpore

Preprocessing and analysis of MRI and fMRI imaging using Freesurfer, FSL, SPM12 and the CONN toolbox, applying PySpark and parallel processing to distribute preprocessing across large medical imaging datasets.

I also built the high-performance workstation the lab ran it on, which cut turnaround on the research workflow substantially.

Freesurfer, FSL, SPM12, CONN Toolbox, Apache PySpark

Investigated functional connectivity in Parkinson's Disease patients against healthy controls using the publicly available Tao Wu dataset, analysing MRI and fMRI data in SPM12, MATLAB and the CONN toolbox with a focus on the cerebellum. Used Apache PySpark to parallelise preprocessing across the dataset.

MATLAB, SPM12, CONN Toolbox, Apache PySpark