+:+===**++;;:^^^:=***===+;`^:;;:::;::::^ arsalan@github
=:===##*+;;;:^^^:;:^`````^`^;;;:::;::::^ --------------------------------
;:=*=#%**=;:`'. . .`^:;;;;;::^: Role......: AI Performance Engineer
:;**=##=*=` ....`^:^^```. '^;;:;:::: Company...: ZAL Analytics (@ZAL-Analytics)
:;=*+===+`';=*##%%@@@@@%%%#=:. `:::;;:: Location..: Utrecht Area, Netherlands
^;==;;++`:%@%%@@@@@@@@@@@%%%%*' ^;;:::; Uptime....: 5+ years shipping AI systems
`+==;++:'*%%%%@@@@@@@@@@@@@%%%:. ';:::::
^+==;++^^#%@@@@@@@@@@%#*===#%%;`'.:::::: Kernel....: Rust (+ C / C++ / CUDA / HIP)
^=*=:^^^:#*;:::;#@@@+``^::^^:#*:`':::::; Runtime...: Vulkan · TensorRT · ONNX · llama.cpp
^==+^^^^::^;++;;=@@@%*=;:;=*##%=::::::^: Focus.....: Low-bit inference · GPU · Air-gapped AI
^+=;^^^:;=*++'';%%@@@%%=``**%%%*=*=;::^^ Stack.....: LoRA · Ontology · RAG · AI Agents
:+++;;;;+=*#%==%@%%%%%@@%%%%%%%==**;::^^
+==*****==*%%%@@%#%%%@@@@@%%%%#+**+:^^^: Repos.....: 34 public (100+ stars, 1k+ downloads)
+++===++++*%%%@@%=**=#%%@@%%%%=;#*^:^^^^ History...: ex Tether · Nscale · AIM Sport · AMD · NXP
=***#*==+++*%%%=:`'`^^^^:*%%#=:^;::::^^^ Edu.......: MSc Data Science · BSc Tech Informatics
****==+=++;^+#;^;=**#***;^*+:^`:::::^^^^ Langs.....: EN (pro) · NL (native) · FR · DE
***=+==*=++;'``^*#+^:+*=:'`''^=*%;:^^^^^ Editor....: nvim · vscode · zed
+===+==*=++=;`..'^`'````...':*##%+^::^::
;++;:;=*=+++**`'....'....':+###*##+::^:: Contact...: arsalan@anwari.nl
:::^^;;=++++*=:::^``''^:;+*###**#=*#==;; LinkedIn..: in/arsalan-anwari
`''`::;::;+=*=;;;;;+++++=#####=++**#%%%#
``'``^:+*#%#*+;;;+;;;++*#**=+++=**####%% Colors....: @@@ ### %%% *** +++ === ::: ...
^:;+*%%%%##*=;::;;;;+;++;;;+==*#########
*#@%%######**+;;:::::;;;+=***######*####
###**####*******=====*******#****#******
AI Engineer with a background in low power and low bit inference, squeezing models onto whatever silicon is in front of me optimized for latency or power usage, from GB200 racks down to low-power embedded boards. Also someone who builds the layer between raw data and AI that reasons over it. Entity resolution, knowledge graphs, RAG and MCP tooling, plus the pipelines, CI/CD and infrastructure to run them in production. Comfortable sitting with domain experts and turning domain knowledge into explicit, testable definitions.
Unlike other AI engineers I care about the parts most people skip: profiling, kernels, portability across heterogeneous back-ends, and making things reproducible.
- Comfortable across the stack: Rust · C · C++ · CUDA · HIP/ROCm · Vulkan · Python
- Moving to Japan in 2027/2028 for work.
Dutch citizen, 5+ years working across NL, CH, FR, BE and NO. Relocating to Japan long term.
Taking JLPT N5 exam in December 2026, N4 in July 2027. Aiming for N2 by end of 2028. Also assisting with building open source eLearning applications for JLPT training.
Languages
GPU / Inference
