# Carla HQ Technical Journal & Research Notes

> Technical deep-dives into tabular foundation models, in-browser WebGPU execution, explainable AI, and zero-shot forecasting.

## Articles

### [Time series forecasting in Google Sheets with Timestral and WebGPU](/blog/time-series-forecasting-in-the-browser-with-timestral.md)
*Published Oct 26, 2026 by Daniel Münch (Timestral, Time Series, WebGPU, Google Sheets, Forecasting, Tabular AI)*

Run Timestral-123M directly in Google Sheets using WebGPU. Discover how Carla brings private, probabilistic time series forecasting to your browser.

- [Read Markdown](/blog/time-series-forecasting-in-the-browser-with-timestral.md) · [Read HTML Article](https://carlahq.eu/blog/time-series-forecasting-in-the-browser-with-timestral)

### [Jev and System One Models for Tabular Data: Semantic Features for TabPFN and TabICL](/blog/encoding-strings-in-tabular-foundation-models.md)
*Published Sep 28, 2026 by Daniel Münch (Jev, System One, TabPFN, TabICLv2, Feature Engineering, Embeddings)*

Use Jev and System One models to turn free-form text into probabilistic semantic features for TabPFN, TabICL and other tabular foundation models.

- [Read Markdown](/blog/encoding-strings-in-tabular-foundation-models.md) · [Read HTML Article](https://carlahq.eu/blog/encoding-strings-in-tabular-foundation-models)

### [TabICLv2 vs TabPFN-3 vs TabPFN-2.6 vs TabFM vs EXAONE: which tabular foundation model can run in the browser?](/blog/tabiclv2-vs-tabpfn3-vs-tabpfn26-vs-tabfm-vs-exaone-which-tabular-foundation-model-can-run-in-the-browser.md)
*Published Aug 13, 2026 by Daniel Münch (TabPFN-3, TabPFN-2.6, TabICL, TabICLv2, Benchmark)*

Evaluating TabICLv2, TabPFN-3, TabFM, and EXAONE-Tabular on model weight size, latency, licensing, and WebGPU browser execution compatibility.

- [Read Markdown](/blog/tabiclv2-vs-tabpfn3-vs-tabpfn26-vs-tabfm-vs-exaone-which-tabular-foundation-model-can-run-in-the-browser.md) · [Read HTML Article](https://carlahq.eu/blog/tabiclv2-vs-tabpfn3-vs-tabpfn26-vs-tabfm-vs-exaone-which-tabular-foundation-model-can-run-in-the-browser)

### [Demystifying SHAP: How Game Theory Explains Machine Learning Predictions](/blog/demystifying-shap-explainable-ai.md)
*Published Aug 11, 2026 by Daniel Münch (ExplainableAI, SHAP, MachineLearning, Carla)*

What explainable AI is, how Shapley Additive exPlanations work from first principles, and how to read SHAP waterfall and directional beeswarm plots.

- [Read Markdown](/blog/demystifying-shap-explainable-ai.md) · [Read HTML Article](https://carlahq.eu/blog/demystifying-shap-explainable-ai)

### [Generating Privacy-Preserving Synthetic Tabular Data in Google Sheets](/blog/synthetic-data-generation-dp-sgld.md)
*Published Aug 8, 2026 by Daniel Münch (SyntheticData, DifferentialPrivacy, Google Sheets, Carla)*

How Carla uses Langevin diffusion and differential privacy to generate privacy-preserving synthetic data directly inside your spreadsheets.

- [Read Markdown](/blog/synthetic-data-generation-dp-sgld.md) · [Read HTML Article](https://carlahq.eu/blog/synthetic-data-generation-dp-sgld)

### [Escaping Google's OAuth Purgatory for a Local AI Chrome Extension](/blog/oauth-consent-screen-approval.md)
*Published Aug 6, 2026 by Daniel Münch (OAuth, Google Sheets, Security, BuildingInPublic)*

How we navigated Google Cloud Trust & Safety OAuth verification for Carla.

- [Read Markdown](/blog/oauth-consent-screen-approval.md) · [Read HTML Article](https://carlahq.eu/blog/oauth-consent-screen-approval)