AI engineer · Industrial systems
I'm Serdar Gündoğdu—an AI engineer who turns complex industrial data into usable products, dependable machine learning systems, and faster decisions.
The work behind me is a traceable journey from physical systems to AI.
ActiveA dated view of where my attention and work are going now.
HorizonThe long-term direction is open humanoid engineering in the physical world.
PublishedOnly explicitly approved public memory snapshots appear here.
Working outputsAccumulated knowledge becomes focused applications, labs, and long-term work.
Career journey
I started with mechanics; industrial engineering and a materials-and-manufacturing master's taught me to see systems through a multidisciplinary, research-led lens. As a production engineer and manager, I worked with lean flows, quality, Six Sigma, and leadership. I then combined that industrial foundation with data science, software, and cloud engineering to build AI products — today I design and deliver AI systems with Claude and Codex as engineering partners.
Now · Primary focus
I design and build AI systems end to end, connecting model development, dataset engineering, evaluation, software, and cloud delivery. My current work also includes model adaptation, agentic workflows, and evidence-aware evaluation.
Product engineering
Connected models to complete products through frontend, backend, APIs, deployment, and cloud engineering.
Data
Moved from process evidence to models, algorithms, data pipelines, and analytical software.
Leadership
Led teams and processes with a focus on quality, Six Sigma, repeatability, and continuous improvement.
Operations
Translated engineering discipline into stable production flow, value stream mapping, and Lean improvement.
Graduate study
Strengthened an evidence-driven approach through materials, manufacturing, and academic research.
Education
Learned to connect people, processes, technology, and decisions as one operating system.
Education
Built a foundation in mechanics, physical systems, and engineering problem solving.
Learning system · AI Ecosystem Atlas
The ten applications form one learning loop: AIA orients the system; GPU provides the compute foundation; USL adapts models; and LLM connects both through runtime and serving. HNS maps the harness layer that turns served model capability into a reliable agent system; CTX designs what the model actually sees and feeds that context back into the LLM runtime while also informing the local-compute path in LCL; SEC then tests whether that system can be trusted from identity through audit and incident recovery; EVL closes the loop with evidence rather than assumption. SEC informs both LCL and CLD, while EVL informs the cloud-services path in CLD. Below the loop, the horizon spine descends from LCL and CLD through WFM, ITL, and ENG — world models, industrial twins, and the long-term embodied destination the whole learning system serves.
Learning path · foundations first, agent systems then assurance, deployment decisions in parallel
aiaLiving map · never “done”aiaLiving map · never “done”opens in a new tabgpuFoundationgpuFoundationopens in a new tabllmMain projectllmMain projectopens in a new tabuslFrom running to changing modelsuslFrom running to changing modelsopens in a new tabhnsReliable agent-system layerhnsReliable agent-system layeropens in a new tabctxWhat the model actually seesctxWhat the model actually seesopens in a new tabsecTrust and assurance layersecTrust and assurance layeropens in a new tabevlEvidence over assumptionevlEvidence over assumptionopens in a new tablclLocal deployment decisionlclLocal deployment decisionopens in a new tabcldCloud deployment decisioncldCloud deployment decisionopens in a new tabOrient
aia“What exists?”
Start with the map of the territory. Compare AI products and developer ecosystems with evidence before investing time in any track. The atlas is a living index, not a project to finish — every run through the loop updates it.
ecosystem models training inference runtime hardware cloud
aia.aserdargun.com opens in a new tabBuild · three parallel tracks
gpu“How does compute work?”
GPU Kernel Engineering — Kernel Atlas
Learn the machine the models run on: CUDA, Triton, GPU memory, profiling, and kernel optimization across a bilingual 12-week atlas.
CPU vs GPU GPU architecture VRAM memory bandwidth CUDA / Tensor cores FP32 / FP16 / BF16 / FP8 / INT8 / INT4 matrix multiplication CUDA kernels FlashAttention KV cache quantization multi-GPU tensor parallelism
gpu.aserdargun.com opens in a new tabllm“How do models run?”
LLM Runtime & Serving Atlas
Learn to ship models: runtime and serving solutions classified across seven architectural layers. Hardware and training both feed this track.
model architecture precision memory calculator runtime inference engine serving API benchmark
Ollama llama.cpp vLLM SGLang TensorRT-LLM Transformers MLX
llm.aserdargun.com opens in a new tabusl“How do models learn/change?”
Unsloth Studio Learning
Learn to change models, not just run them: Unsloth, LoRA, QLoRA, dataset engineering, and evaluation in a bilingual learning atlas.
pretrained model dataset tokenization LoRA QLoRA SFT DPO GRPO evaluation merged model LLM runtime
usl.aserdargun.com opens in a new tabgpu → llm hardware decides what you can serve usl → llm the models you train become what you serve llm → gpu inference pressure exposes the hardware limits
Engineer · agent system
hns“How do I turn model capability into a reliable agent system?”
Harness Engineering Observatory
Turn a served model into a dependable agent system. Compare harnesses, frameworks, runtimes, orchestration, sandboxes, verification, and observability with explicit evidence, freshness, and tradeoffs before choosing the surrounding system.
model capability context tools orchestration sandbox memory verification observability reliable agent system
hns.aserdargun.com opens in a new tabEngineer · context and assurance
ctx“What does the model actually see?”
Context Engineering Observatory
Design the prompt and everything around it. Trace the system message, context window, chunking, embeddings, vector search, retrieval, memory, and tool use so the model is only as good as the context in front of it.
system message context window chunking embeddings vector DB retrieval memory tool use token budget
ctx.aserdargun.com opens in a new tabsec“Why should I trust this agent system?”
AI Systems Security Observatory
Trace delegated intent from model capability to auditable action. Examine identity, credentials, authorization, tool and sandbox boundaries, protected data, audit evidence, and incident recovery without collapsing assurance into a score.
model agent identity credential authorization tool sandbox data action audit incident
sec.aserdargun.com opens in a new tabevl“How do I know it works?”
LLM Evaluation Observatory
Close the loop with evidence. Build golden sets, choose metrics, run LLM-as-judge, A/B tests, regression suites, and online monitoring so every release is a measured change, not an assumption.
golden set metrics LLM-as-judge A/B test regression suite human eval benchmark online monitoring
evl.aserdargun.com opens in a new tabctx → sec context decides what is allowed in, security decides what is allowed out sec → evl assurance surfaces the questions evaluation must answer evl → ctx measurement feeds the next round of context design
Deploy · local and cloud in parallel
lcl“Which local lab should I buy?”
Local Compute Lab
Match open-weight model requirements to NVIDIA, AMD, and Apple devices. Compare workload, memory, privacy, power, noise, and existing equipment before choosing a local lab.
open-weight model workload memory NVIDIA / AMD / Apple privacy / power / noise local lab
lcl.aserdargun.com opens in a new tabcld“How do I operate this at scale?”
Cloud Provider Cost Comparison
Turn the stack into decisions: compare pre-tax USD costs of cloud services available from Türkiye and choose where the whole system runs. Cloud comes last on purpose — once you know what you deploy, infrastructure stops being abstract.
model vLLM Docker GPU instance cloud GPU load balancer autoscaling API
cld.aserdargun.com opens in a new tabReturn
The loop closes when what you learned returns to the atlas — the map you started with becomes the record of what you can now build, and the next pass through the loop starts deeper.
From physical systems
to governed intelligence
to decisions people can trust
Application map · live destinations
Ten core learning applications, one lab, one horizon bridge, and one long-term horizon. The three-letter code is the permanent key between each application, repository, and aserdargun.com subdomain.
| Code | Application | Repository | Address |
|---|---|---|---|
aia | AI Ecosystem AtlasAn evidence-backed research console for comparing AI products and developer ecosystems.Review soon | aia-aserdargun-com opens in a new tab | aia.aserdargun.com opens in a new tab |
llm | LLM Runtime & Serving AtlasA source-backed field guide to LLM runtime and serving solutions across seven architectural layers.Review soon | llm-aserdargun-com opens in a new tab | llm.aserdargun.com opens in a new tab |
hns | Harness Engineering ObservatoryA bilingual, source-backed observatory for comparing the harnesses, runtimes, orchestration, execution, verification, and observability layers that turn model capability into reliable agent systems.Current | hns-aserdargun-com opens in a new tab | hns.aserdargun.com opens in a new tab |
sec | AI Systems Security ObservatoryA bilingual, evidence-aware observatory for tracing AI-agent trust from model intent through identity, authorization, constrained action, audit, and incident recovery.Current | sec-aserdargun-com opens in a new tab | sec.aserdargun.com opens in a new tab |
ctx | Context Engineering ObservatoryA bilingual observatory for the prompt, retrieval, and memory decisions that decide what the model actually sees — system messages, chunking, embeddings, vector search, tool use, and token budgets tracked against evidence.Current | ctx-aserdargun-com opens in a new tab | ctx.aserdargun.com opens in a new tab |
evl | LLM Evaluation ObservatoryA bilingual observatory for LLM evaluation: golden sets, metrics, LLM-as-judge, A/B tests, regression suites, and online monitoring that turn releases into evidence rather than assumptions.Current | evl-aserdargun-com opens in a new tab | evl.aserdargun.com opens in a new tab |
usl | Unsloth Studio LearningA bilingual learning atlas for Unsloth, LoRA, QLoRA, dataset engineering, and evaluation.Review soon | usl-aserdargun-com opens in a new tab | usl.aserdargun.com opens in a new tab |
gpu | GPU Kernel Engineering — Kernel AtlasA bilingual 12-week atlas for CUDA, Triton, GPU memory, profiling, and kernel optimization.Review soon | gpu-aserdargun-com opens in a new tab | gpu.aserdargun.com opens in a new tab |
cld | Cloud Provider Cost ComparisonA source-backed decision tool comparing pre-tax USD costs for cloud services available from Türkiye.Review soon | cld-aserdargun-com opens in a new tab | cld.aserdargun.com opens in a new tab |
lcl | Local Compute LabA source-backed decision workbench matching open-weight model requirements with NVIDIA, AMD, and Apple devices for a local AI lab.Current | lcl-aserdargun-com opens in a new tab | lcl.aserdargun.com opens in a new tab |
wfm | World Models AtlasA living research atlas tracing how world models connect perception, prediction, planning, and action through primary sources.Current | wfm-aserdargun-com opens in a new tab | wfm.aserdargun.com opens in a new tab |
itl | Industrial Twin LabAn isolated experimentation environment for digital twins, industrial AI, simulation, and evidence-driven machine knowledge — where the AI stack meets the physical asset.Review soon | itl-aserdargun-com opens in a new tab | itl.aserdargun.com opens in a new tab |
eng | Open Humanoid EngineeringThe long-term horizon for embodied, open humanoid engineering — what the atlas, hardware, runtime, training, cloud, and industrial twin lab all serve.Horizon · in design | eng-aserdargun-com opens in a new tab | eng.aserdargun.com opens in a new tab |
How I work
I combine industrial context, statistical thinking, and modern AI development workflows to close the gap between an idea and a system people can actually use.
Start with the operator, engineer, or business decision—not the model. Define the signal, constraint, and useful action.
Connect data, analysis, APIs, interface, and feedback so the intelligence survives outside a notebook.
Design for traceability, human review, deployment constraints, and the realities of industrial environments.
About
My background spans statistics, mathematics, mechanical and industrial engineering. That mix helps me see AI as a complete operating system: the data foundation, the model, the interface, the deployment path, and the people making the final decision.
I use AI coding partners such as Codex, Claude Code, and Gemini as part of a disciplined development workflow—accelerating research and implementation while keeping architecture, validation, and accountability human-led.
Open to remote collaborations · TR / EN
Investment · depth beats order
The study order is AIA → GPU → LLM → USL → CLD; the depth priority is LLM first. Right now the center of gravity is the GPU + LLM pair — where the hardware meets the runtime. The horizon spine (WFM, ITL, ENG) is the long-term destination, kept visible at 1% each so the loop never forgets what it serves.