Steffen Hirschmann
Senior AI Training Inference Performance Engineer at Aleph Alpha
Stuttgart Region Germany
Summary
👤
Senior🎓
Top SchoolSteffen Hirschmann is a Senior AI Training & Inference Performance Engineer with 10 years of experience optimizing latency and throughput across CPUs, GPUs, DSPs and NPUs for edge and data-center workloads. He has driven concrete wins—from doubling MoE throughput in vLLM on an 8× H200 cluster to a 30× latency reduction for a Qualcomm Hexagon NPU CNN head and a 10× speed-up porting radar algorithms to TI C66 DSPs—while cutting memory footprints and hardware costs in ADAS projects. Skilled in low-level profiling and runtime tuning (Linux Perf, VTune, Nsight), modern C++/CUDA, assembly and embedded NN stacks, he also builds cross-platform porting paths between CPU, GPU, DSP and NPU. A PhD-trained researcher and co-founder who scaled both scientific MPI projects (ESPResSo contributions) and an edge-AI optimization startup, he regularly mentors peers and has run workshops to disseminate performance best practices. Notably, he implemented MPI-IO support in ESPResSo and created an x86 AVX2 emulator for TI DSP intrinsics to speed developer iteration, blending deep research experience with hands-on production impact.
10 years of coding experience
6 years of employment as a software developer
Doctor of Science (PhD), Computer Science, Doctor of Science (PhD), Computer Science at University of Stuttgart
Github Skills (9)
mpi10
c-language10
molecular-dynamics-simulation10
cprogramming-language10
molecular-simulation10
nonlinear-dynamics10
scientific-computing10
python9
tcl9
Programming languages (10)
Github contributions (5)
espressomd/espresso
Jan 2016 - Apr 2020
The ESPResSo package
Role in this project:
Back-end Developer
Back-end DeveloperContributions:134 commits, 37 PRs, 62 comments in 4 years 3 months
Contributions summary:Steffen primarily contributed to the ESPResSo package by implementing and integrating functionality for MPI-IO, a parallel I/O library, enabling the writing and reading of particle data. They implemented Tcl wrappers for MPI-IO, added a Python script for converting MPI-IO files, and addressed issues related to the functionality and file handling. Further contributions involved minor code fixes and improvements, such as fixing output formatting and correcting an issue related to a dangling statement.
physicssoft-matterpythonmolecular-dynamicsscientific-computing
hirschsn/espresso
Jan 2016 - May 2021
The ESPResSo package
Contributions:125 pushes, 67 branches in 5 years 4 months
espressoesp32
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.