Serdar Gündoğdu

AI Learning System

Learn AI. Put it to work.

Explore how AI systems work through research, applications, and hands-on experiments. Follow five connected layers from compute and models to agents, deployment, and physical AI. Connect ideas, test what you learn, and use it to build your next project.

Scroll down to explore the five layers →

01

Foundation

Ecosystem, compute, runtime, and model-building foundations, with GEX for hands-on GPU execution.

05

Physical AI

World models and swarm research meet WML and ANT / BEE experiments, PDT pump twins, and HEX humanoid exploration.

Indented applications belong to the application above them.

AI Learning System · connected paths

How the system connects.

Each application explores a different part of building AI. The diagram connects them into a learning loop: understand compute and models, build agent systems, and test how they behave. Compare local and cloud deployment, then explore world models, swarm intelligence, and colony experiments. These paths inform digital twins and the longer-term goal of open humanoid engineering. Each experiment brings new questions and lessons back to the starting point.

Pinch with two fingers to zoom. Drag to explore the enlarged map.

Connected applications and their sub-applications Large boxes show the connected main applications. Smaller boxes are sub-applications, enclosed with their parent in a shared frame. AIA connects GPU and USL to LLM, then HNS, CTX, SEC and EVL. CTX feeds back to LLM. Assurance informs parallel LCL and CLD deployment; both connect to WFM and SWI, then ITL and ENG. GPU owns GEX, WFM owns WML, SWI owns ANT and BEE, ITL owns PDT, and ENG owns HEX. A dot marks the shared deployment junction. Opens in a new tab AIA AI ecosystem Opens in a new tab GPU GPU kernels Opens in a new tab GEX Kernel execution Opens in a new tab LLM Runtime · serving Opens in a new tab TFL Token serving Opens in a new tab USL Model training Opens in a new tab ADP Model adaptation Opens in a new tab HNS Harness engineering Opens in a new tab ARL Agent runtime Opens in a new tab CTX Context · knowledge Opens in a new tab SEC Trust · security Opens in a new tab EVL Evaluation Opens in a new tab LCL Local compute Opens in a new tab CLD Cloud deployment Opens in a new tab WFM World models Opens in a new tab WML Prediction experiments Opens in a new tab SWI Swarm intelligence Opens in a new tab ANT Pheromone trails Opens in a new tab BEE Dance communication Opens in a new tab ITL Industrial twins Opens in a new tab PDT Pump digital twin Opens in a new tab ENG Humanoid engineering Opens in a new tab HEX Humanoid exploration
Follow the arrows between main applications. The smaller boxes share a frame with their parent application.

From research to hands-on learning

Explore an idea. Run an experiment.

Follow each research application into its companion learning experience. These links describe learning relationships; each application runs independently.

LLM → TFLLive

How does a request become a token?

A deterministic, bilingual educational laboratory that follows individual LLM requests through queueing, admission, chunked prefill, KV state, iterative decode, sampling, and token delivery, exploring competing request lifetimes, conservative memory reservations, static versus continuous batching, and latency/throughput trade-offs.

Research · LLM ↗ opens in a new tab Explore · TFL ↗ opens in a new tab

HNS → ARLLive

What happens between intent and action?

A deterministic, bilingual educational agent runtime that inspects a single revenue-update execution across HNS (Harness), CTX (Context), SEC (Security), and EVL (Evaluation). All model decisions are scripted; no LLM API, no live data, no remote telemetry.

Research · HNS ↗ opens in a new tab Explore · ARL ↗ opens in a new tab

USL → ADPLive

How does a base model become adapted to a task?

A bilingual, browser-only educational laboratory for Full FT, LoRA and QLoRA. The conceptual companion to USL: USL teaches the adaptation landscape; ADP makes the trade-offs inspectable through interactive parameter and memory footprints.

Research · USL ↗ opens in a new tab Explore · ADP ↗ opens in a new tab

Explore the journey