Loading

I build observable factories.

Engineer and researcher at KIT, working toward industrial system architecture.

My work starts where most industrial AI cannot: factories with limited digital infrastructure. I design the sensing, data, and software layers that make production observable, maintainable, and ready for adaptive inference.

Explore

01 · observability

Minimum viable observability

Before a factory can become AI-native, it needs reliable traces of what is happening. I focus on the smallest useful layer of sensors, manual checkpoints, event streams, and process records that can make a production line observable.

The first milestone is not automation. It is knowing what is happening well enough to reason about it.

02 · backbone

Reliable industrial data infrastructure

Low-digital-maturity factories rarely need another dashboard first. They need timestamps, identifiers, schemas, validation, storage, and ownership: the quiet infrastructure that turns raw activity into usable data.

A factory that owns its data can improve its process without renting its memory from disconnected tools.

03 · architecture

Systems that can be maintained

I am deliberately moving toward system architecture and long-term maintainership: designing software, infrastructure, and operating routines that a manufacturing company can keep using after the pilot is over.

The work has to survive contact with real operators, real orders, and real maintenance constraints.

04 · inference

Uncertainty-aware production models

Once the factory is observable, the research opens up: process reconstruction, probabilistic state estimation, belief propagation, adaptive inference, world models, and decision support for incomplete industrial data.

Advanced AI becomes meaningful only after the production system has a structured reality to reason over.

05 · practice

Research carried by real factories

KIT gives the academic environment for my researchs; motivated facilities gives a real manufacturing validation environment. The company work is the vehicle that tests whether the methods survive practice.

Research creates the technology, pilots validate it, and the company carries it into industry.

01 / 05

Positioning

Engineer, researcher, system architect in training.

I work at the boundary between industrial software engineering and applied research. My focus is how low-digital-maturity manufacturing environments can become observable, data-driven, and eventually adaptive without pretending they already have clean real-time data.

  • KITThe academic frame is methodology: how to transform manual factories into observable production systems.
  • practiceThe practical frame is maintainable infrastructure: sensors, data pipelines, software, and operating routines that real teams can use.

How it starts

Start with one production line.

The first research pilot should be narrow enough to validate. One workflow, one line, one set of events. Map the process, instrument what matters, capture missing and noisy data honestly, then reconstruct the state of production.

  • 01 mapDefine the real process: jobs, machines, operators, materials, checkpoints, delays, and failure modes.
  • 02 observeBuild the minimum viable data layer with sensors, manual inputs, timestamps, IDs, and validation.
  • 03 inferUse the resulting traces for analytics, monitoring, probabilistic state estimation, and eventually decision support.

Long arc

From paper workflows to adaptive industrial systems.

The long-term goal is not a flashy AI layer on top of chaos. It is a maintained system architecture where data collection, process knowledge, uncertainty-aware inference, and industrial agents can grow from the same foundation.

  • nowBuild observability and reliable data capture for low-digital-maturity manufacturing.
  • nextDevelop monitoring, process analytics, and uncertainty-aware models on real factory traces.
  • laterMove toward adaptive decision support, world models, and industrial agents with a real operating substrate.
Validation Environment

Batumi Socks | Skytex Georgia

active

Skytex Georgia is a real embroidery manufacturing environment with low digital infrastructure and mostly manual workflows. That makes it a strong validation environment for my research goal: turning a factory from paper-based operation into an observable, maintainable, and eventually AI-native production system. The current work combines product and wholesale software with the first owned digital operating layer.

1 lineminimum viable observability target
15 scustomer idea to AI-generated product concept
0 → 1from paper workflows toward owned data infrastructure

Research and architecture

next

I am looking for supervisors, collaborators, and manufacturing environments where this work can mature from pilot infrastructure into maintainable industrial systems. Write me if this overlaps with your work.

© 2026 Alper Sarıtaş