Patok Lab
Join Early Access

Early access · macOS · Windows · Linux

Run fewer experiments. Learn faster.

Patok Lab is a local AI workspace for experimental R&D — bringing experimental data, surrogate models, sequential optimization, scientific literature and Python-powered AI into one research project.

Join Early Access Bring Patok Lab to your research group
Patok Lab — Explore → Surface matrix: every factor pair of the Cold Spray campaign drawn as a contour of the fitted surrogate, sliced through the best observed run.

The Cold Spray example campaign in Patok Lab: pressure, temperature and stand-off distance against deposition efficiency, sliced through the best run.

Born from Cold Spray researchScientific Python bundledLocal AI modelsSix example projectsNo telemetry

The problem

Experimental research is fragmented.

The data lives in a spreadsheet, the model in a script, the reasoning in a notebook, the papers in a folder and the AI conversation somewhere else. Six months later nobody can say what the campaign knew at run twelve.

  • Excel
  • Python scripts
  • Papers
  • Notes
  • Plots
  • Optimization code
  • Separate AI chats

One research project

Patok Lab keeps the experimental campaign together.

A project is a folder of plain files — data, models, episodes, notes, literature and reports — with provenance from the first run to the last recommendation.

  • Data — experimental reality, versioned
  • Models — surrogates you can verify
  • Episodes — recommend → run → observe → update
  • Notebook — the researcher’s thinking
  • Literature — references and PDFs, indexed locally
  • Artifacts — reports with provenance

The workflow

From a table of runs to the next experiment.

Six steps, one project. Every screen below is the real application.

  1. 01

    Bring your experiments

    Import Excel or CSV. Pick the sheet and header row, classify columns as inputs, responses, categorical factors and metadata, set units and physical bounds. Add formula columns and one-click transformations; the app suggests the next analysis from the data itself. The source file is copied read-only and every save is a new version.

    • XLSX / CSV
    • Column roles & bounds
    • ƒ formula columns
    • Versioned saves
    Patok Lab Runs view: the Cold Spray campaign table with column headers colored by role — inputs, response and metadata.
  2. 02

    Understand what happened

    Scatter, matrix, distributions, correlation and parallel coordinates; the response surface of the active model over any two factors with the rest pinned by sliders; the surface matrix, every factor pair at once. Statistics and curve fitting are one click away from any column or selection.

    • 8 exploration views
    • Response surfaces
    • Statistics & curve fit
    • Export for publication
    Patok Lab Explore view: response surface of deposition efficiency over pressure and temperature with stand-off distance pinned by a slider.
  3. 03

    Model the process

    Fit polynomial, Gaussian-process, random-forest and other surrogates — 18 regression and 9 classification types, categorical factors included. Verify with cross-validation, held-out tests, residuals, learning curves and permutation importance, and get an honest verdict. Compare every model type in one ranked table.

    • 18 + 9 model types
    • Verification verdict
    • Compare all
    • Predict from a file
    Patok Lab Models view: verification report for the fitted surrogate with predicted-versus-observed, residuals and learning-curve diagnostics.
  4. 04

    Decide what to run next

    Choose objectives — maximize, minimize or hit a target — with bounds, frozen factors and constraints. The validated sequential optimizer recommends the next experiment, with the predicted response and where it sits on the surrogate.

    • Objectives & constraints
    • Seeded, reproducible
    • Promote to Episode
    Patok Lab Optimize view: a recommended next experiment card with predicted deposition efficiency, above the surrogate slices where the recommendation sits.
  5. 05

    Learn from every episode

    Promote a recommendation to an Episode, record what was performed and what was observed, and the campaign updates. Step back to any episode and see what the model knew at that point — and what it did not.

    • Recommend → run → observe
    • History with foresight
    • Provenance
    Patok Lab Episodes view: sequential history of predicted versus observed deposition efficiency, with the recommended, performed and observed panels of episode one.
  6. 06

    Research with AI

    Ask the research agent. It runs the same recommender as the button, analyzes the table with pandas, proposes computed columns with their assumptions, searches your papers with page provenance and builds HTML reports. Citations that do not resolve to a saved reference are rejected by a deterministic guard. Every tool call is visible; edits need your confirmation. Runs with a local model.

    • Python & pandas
    • Proposals, not surprises
    • Citations that must resolve
    Patok Lab research agent panel next to a response surface: the agent ran the validated engine and proposes the next experiment with predicted deposition efficiency.

The toolkit

Everything a table of runs needs, in one place.

Under the sequential loop sits a complete analysis workbench. The science is scipy and scikit-learn; what Patok Lab adds is the workflow: roles and bounds, versions, provenance, and the same three actions everywhere — Save as artifact, Export for publication, Ask the agent.

  • Patok Lab Statistics view: an I-MR control chart of coating thickness with control limits, rule violations table and a plain-language reading.

    Statistics — I-MR control chart with Western Electric rules and a plain-language reading

  • Patok Lab Explore → Curve fit: thirteen function shapes ranked by AIC for reaction-rate data, with parameters, standard errors and confidence intervals.

    Curve fitting — Thirteen shapes ranked by AIC on Arrhenius kinetics data

  • Patok Lab Design (DOE) → Analyze this table: effects table, Pareto plot and curvature check for a fractional factorial sintering design.

    Design of experiments — Effects analysis of a 2⁴⁻¹ screening design with center points

  • Statistics

    Fourteen procedures: descriptive and normality, t-tests, one-way ANOVA with Tukey, non-parametric tests, chi-square, correlation, process capability, I-MR and X̄-R/S control charts, power and sample size, Gauge R&R. Each returns tables, figures and a reading in plain language.

  • Curve fitting

    Thirteen physically motivated functions plus your own expression, parameters with standard errors and confidence intervals, prediction bands, residuals, and a ranking of every shape by AIC. Add fitted values back as a column.

  • Design of experiments

    Full and fractional factorials, Plackett-Burman, central composite, Box-Behnken and Latin hypercube designs in the classic layout; effects, Pareto and half-normal plots, aliasing and curvature for the runs you already have; append planned runs to the table.

  • Model zoo, verified

    Ridge, lasso, PLS, SVR, kernel ridge, random forest, gradient boosting, neural nets and their classifiers — all under the same verification protocol, with confusion matrices and ROC for pass/fail responses.

  • Publication figures

    Every chart exports as SVG, PDF or 300-dpi PNG at journal widths, in light or dark theme. A light mode for the whole app when you prepare a manuscript.

  • Literature, notes, artifacts

    References with or without a PDF, each labelled by its evidence (full text · abstract only · reference only); Find state of the art queries open scholarly indexes from your research question; open-access PDFs fetched and indexed locally with page-level search and an in-app viewer. A notebook that cites pages, episodes and artifacts; every saved result re-runnable from its provenance.

Why this is different

Built around the experimental campaign.

  • The experiment is the project

    Patok Lab understands inputs, responses, models, recommendations, episodes and observed results — not just a file. A project is a folder of plain files you own.

  • From analysis to action

    It does not stop at “what happened?”. Define objectives and constraints and it helps answer “what should I test next?” — with a validated sequential optimizer, not a chart.

  • AI that can actually analyze

    The research agent runs scientific Python and pandas on your data and calls validated scientific tools. It operates the same services as the buttons, and shows its work.

  • Local by design

    The app, the scientific Python runtime and the local model server run on your machine. Research can stay on your laptop, and local models run without sending project data to the cloud.

Flagship scientific case

Where Patok Lab comes from: Cold Spray.

The sequential-experimentation engine inside Patok Lab was developed on a Cold Spray deposition-efficiency campaign for austenitic stainless steel. Six screening runs, then five active-learning episodes — each recommended by a surrogate model and run with two stand-off replicates — reached approximately 97% deposition efficiency.

That campaign ships as a one-click example project. The engine is generic: any table of inputs and measured responses. Cold Spray is the reference case, not the boundary.

  • 6
    screening runs
  • 5 × 2
    active-learning episodes × replicates
  • ≈97%
    deposition efficiency reached
  • 33.8 bar · 750 °C · 20 mm
    pressure · temperature · stand-off
Patok Lab Episodes view on the Cold Spray example: model prediction versus observed deposition efficiency across sixteen runs.

Built for real research

Physical experiments where every run costs time and money.

Patok Lab works on any campaign that can be written as a table of controllable inputs and measured responses. These are the fields we are opening early access to first.

  • Materials science
  • Additive manufacturing
  • Coatings / thermal spray
  • Manufacturing R&D
  • Process development

Who we are opening it to

  • Materials scientists
  • Experimental researchers
  • PhD students & postdocs
  • Principal investigators
  • Research laboratories
  • Industrial R&D engineers

Example projects

Six example projects, one click each.

Each teaches one part of the workflow on a small, realistic dataset. The Cold Spray project is the real published campaign; the others are generated deterministically and ship with hand-out CSVs.

  • Cold Spray campaign — surrogate, response surface, next experiment, episodes
  • Coating adhesion by powder and pressure — ANOVA + Tukey, categorical factors in models
  • Screening design 2⁴⁻¹ + center points — effects analysis, aliasing, curvature
  • Reaction kinetics (Arrhenius) — curve fitting, transformations, publication figures
  • Weld quality pass/fail — classifiers and regression on the same table
  • Coating thickness in production order — control charts and capability
Patok Lab All projects page: the project library and the gallery of six one-click example projects.

Local AI · privacy

Your experiments can stay in your laboratory.

Patok Lab is a desktop application, not a cloud service. What runs where is explicit and inspectable.

  • Patok Lab itself runs locally

    A desktop application for macOS, Windows and Linux. The app, the analysis engine and your project folder — plain CSV, JSON, Markdown and HTML files — live on your machine. Settings → Privacy lists exactly what could ever leave it.

  • Scientific Python is bundled

    numpy, pandas, scikit-learn, scipy and plotly ship inside the app. You never install or manage Python.

  • Local AI models, without a terminal

    Settings → Local models: pick a curated model (Qwen2.5-7B-Instruct is the recommended one), Download → Use. Patok Lab serves it itself with a bundled runtime — no LM Studio, no Ollama, no command line.

  • One connectivity setting, three postures

    Air-gapped: zero connections, local model only, model files imported from disk or USB. Connected research (default): the project and the model stay local; only your literature query text goes to scholarly indexes, and model servers on your lab’s own network are allowed. Cloud-enhanced: additionally a remote AI provider, when you choose a larger model.

  • Bring your own model server

    Any OpenAI-compatible endpoint — on this machine, on your institution’s network, or a hosted API in cloud-enhanced mode. Keys are stored locally, never in the project.

  • No telemetry

    Patok Lab sends no usage data. This is a fact of the build, not a setting.

Patok Lab Settings → Local models: curated model catalog with Download and Use, and Import GGUF from disk or USB.

What we do not claim: OS-level sandboxing. The air-gap guard is policy plus detection, honestly labelled, not kernel-level enforcement. Agent-generated Python runs with your user permissions; mutating actions need your confirmation, and every artifact records how it was produced.

Early access · workshops

We are opening Patok Lab to a small group of experimental researchers and research labs.

Bring a real dataset. We want people using Patok Lab on live experimental campaigns — and telling us what breaks. Early access is free while we learn together.

  • macOS, Windows and Linux — tell us which one you need first
  • Direct line to the founder during early access
  • Optional workshop for your research group: bring a campaign, leave with a next experiment
Current tools optional
Interested in an early-access workshop for your group?

We only use these details to contact you about Patok Lab early access. No newsletters, no resale.

Scientific origin

Made by a researcher who ran the experiments.

Patok Lab was created by Horacio Canales, PhD, from prior work in materials and manufacturing research: Cold Spray process development, surrogate modeling, sequential experimentation and industrial AI.

The sequential-optimization method at the core of the product comes from his Cold Spray research — the campaign that reached about 97% deposition efficiency is the one that ships as the example project. Patok Lab is built by Patok Research, the company behind Patok’s industrial AI products.

Horacio Canales, PhD on LinkedIn Google Scholar

FAQ

Questions researchers ask first.

Does Patok Lab require cloud access?

No. Patok Lab runs on your machine, with scientific Python bundled and a local model server built in. In the default “connected research” posture the only things that go out are a model download you start and, if you use Find state of the art, the text of your literature query to open scholarly indexes. Air-gapped mode blocks everything and you import model files from disk or USB. A remote AI provider is only used if you switch to cloud-enhanced.

Can I use my existing Excel/CSV experiments?

Yes. Import XLSX or CSV, choose the sheet and header row, and classify columns as inputs, responses, categorical factors and metadata with units and physical bounds. The source file is copied read-only; every save is a new version. You can append new runs later from an updated file, paste from Excel, and export the current version back to CSV or XLSX.

Do I need Python installed?

No. A scientific Python runtime — numpy, pandas, scikit-learn, scipy, plotly, matplotlib — ships inside the application. No admin rights are needed. The research agent uses it, and you can view the Python behind any analysis it produces, but you never install or manage Python yourself.

Can I use a local AI model?

Yes. Settings → Local models lists curated models (Qwen2.5-7B-Instruct is the recommended one) with Download → Use, served by the bundled runtime. You can also import a GGUF file, or point Patok Lab at any OpenAI-compatible endpoint such as LM Studio, Ollama, vLLM or a hosted API. The agent needs a model with reliable tool calling; very small models work only partially, and Patok Lab tests this for you when you connect one.

Is Patok Lab only for Cold Spray?

No. Cold Spray is the flagship case and ships as a one-click example, next to five others: powder adhesion (ANOVA), a sintering screening design, Arrhenius kinetics, weld pass/fail and a coating-thickness production log. The engine is generic: any table of controllable inputs and measured responses.

Who is early access for?

Experimental researchers and labs with a real dataset — materials science, additive manufacturing, coatings and thermal spray, manufacturing and process development. Students, PhD students, postdocs, PIs, R&D engineers and lab directors. We are looking for people who will use it on a live campaign and tell us what is missing.

Can you run a workshop for our research group?

Yes. A half-day or full-day session where your group brings its own campaign data, builds the project, fits and verifies models and leaves with a recommended next experiment. Tick “interested in a workshop” in the form and we will get in touch.

What platforms does Patok Lab support?

Patok Lab is a desktop application for macOS, Windows and Linux. Early-access builds are delivered as direct downloads (no app store, no auto-updates), and we ship platforms in the order early-access members ask for them — choose yours in the form.

Is there a trial, and how will licensing work?

Every install runs a 30-day full trial. Licenses activate offline — a short signed key or a file, nothing is sent anywhere — and the direction is perpetual licenses per version for researchers and labs, with optional updates, plus workshops and scientific consulting. When a license or trial ends the app never locks your data: reading, export and backups keep working. Early-access members use Patok Lab free during the pilot.