Skip to content
View kathuman's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report kathuman

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
kathuman/README.md
Daniel Sepulveda Estay, PhD. Mechanical engineer turned supply chain researcher. I build open, tested models that run in the browser, from turbines and heat pumps to networks that perform even when disrupted.

Engineering and supply chain tools that run in the browser. Each one is tested in CI, free to use, and open to inspect.

AI Projects Estay Dynamics LinkedIn ORCID

About

  • Mechanical engineer by training. Civil Mechanical Engineer, Universidad Técnica Federico Santa María (Chile), and Dipl.-Ing. in Mechanical Engineering, DHBW Mosbach (Germany): thermodynamics, turbomachines, heat transfer, CFD and finite elements.
  • Supply chain researcher. MSc in Industrial Engineering, Pontificia Universidad Católica de Chile; SM in Management Science, MIT; PhD in Operations Management, Technical University of Denmark (DTU), including a year as a visiting researcher at the MIT Center for Transportation & Logistics.
  • Across industries. Mining, fast-moving consumer goods (FMCG) and beverage distribution, industrial manufacturing, healthcare, and pharmaceutical manufacturing. Council member of the System Dynamics Society. Based in Copenhagen.

Two disciplines, one way of working

Supply chainMechanical engineering
Global Disruption Atlas
Global Disruption Atlas
How long a network survives a closure, and what recovering faster costs.
Heat Pump Studio
Heat Pump Studio
Design a heat pump from the building's heat loss to the compressor.
Tierwise
Tierwise
Plan capacity across a multi-tier supply chain, with every number proved by the solver.
Turbine Lab
Turbine Lab
Pelton, Francis and Kaplan turbines in 3D, with the physics behind them.
Real Options for Network Design
Real Options for Network Design
What staging, waiting or leasing is worth against one big commitment.
Tube Flow
Tube Flow
Lattice-Boltzmann CFD of flow past a sphere in a tube, validated against theory.
Ripple
Ripple
A supply chain ordering game for classes, with Claude as an optional player.
In the lab: Escapement, Robot Configurator, Plumb, Cobot Lab.
All mechanical engineering tools → (8)
PolicyForge
PolicyForge
Supply chain. In the lab: Supply Pulse, Clockspeed Balancing, Lag Lab, Network Stress Test.
All supply chain tools → (11)

Also on the site: String Art, Tech University Atlas, Epicycles, AI Expectations Gap Map and more.

One result worth a look

Atlas run: daily service level during a 60-day Red Sea closure. With nothing held, service falls to 67% and the event costs $82.0M; with the optimised lever set it stays above 79% and costs $23.3M

On the Atlas sample network, closing the Red Sea for 60 days costs $82.0M with nothing held. The lever set the optimiser recommends costs $18.9M a year, cuts this event to $23.3M, and cuts the expected annual loss from $50.4M to $15.1M, with contracts, standby capacity and air rather than stock. Try it in the Atlas.

What is it worth to stage, wait or lease instead of committing on one forecast?
Real Options run: NPV over 20,000 demand paths. Committing the full footprint now has a mean of $21.2M and a 23.7% chance of a loss; building smaller and expanding in steps has a mean of $26.5M and a 4.1% chance of a loss

In the default case, building smaller and expanding in steps is worth $5.3M more than committing the full footprint now, and cuts the chance of losing money from 24% to 4%. Over 20,000 simulated demand paths, the means match the lattice valuation within $0.3M. Real Options for Network Design.

Why does a commercial move this quarter cost money a year later?
Clockspeed run: when competitors promise faster delivery, matching at once loses 62k against standing still; preparing the supply chain a year earlier and then matching gains 161k, of which 159k comes from moving the two clocks together

When competitors promise faster delivery, matching them at once loses money against standing still. Preparing the supply chain a year earlier and then matching gains 161k: preparing alone is worth 65k, and 159k comes from moving the two clocks in step. Clockspeed Balancing.

Checked before it is used

Atlas tests Tierwise tests Real Options tests Ripple tests Heat Pump Studio tests Turbine Lab tests Tube Flow tests

Each model is tested against something independent of itself: the flow solver against optimality certificates and a second max-flow algorithm, Tierwise's capacity plan against a duality receipt recomputed from the solver's prices, the option lattice against Black–Scholes, the refrigerant cycle against CoolProp, the CFD against Stokes' law, and the Atlas against the PortWatch record. The suites run on every change.

The figures above are runs of each tool's own engine on its built-in example: the Atlas sample company is fictional (real ports and lanes), the Real Options case is its default, and Clockspeed is in model units. The gallery is rebuilt daily from the projects list.

Pinned Loading

  1. claude-projects claude-projects Public

    Claude Projects

    JavaScript

  2. Python-Experiments Python-Experiments Public

    Jupyter Notebook

  3. estay-dynamics estay-dynamics Public

    JavaScript

  4. R-Experiments R-Experiments Public

    Jupyter Notebook

  5. Vensim-Experiments Vensim-Experiments Public

    This repository relates to the development of System Dynamics models in areas such as Public Health, epidemiology and Cyber-resilience.

    Roff 1 1