<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Posts &amp; notes on Kwak Jazz</title><link>https://gwils28.github.io/en/posts/</link><description>Recent content in Posts &amp; notes on Kwak Jazz</description><generator>Hugo -- gohugo.io</generator><language>en</language><copyright>© 2026 Wilson Goma</copyright><atom:link href="https://gwils28.github.io/en/posts/index.xml" rel="self" type="application/rss+xml"/><item><title>Time series foundation models: zero-shot against the specialist</title><link>https://gwils28.github.io/en/posts/tsfm-zero-shot-specialiste/</link><pubDate>Fri, 06 Nov 2026 00:00:00 +0000</pubDate><guid>https://gwils28.github.io/en/posts/tsfm-zero-shot-specialiste/</guid><description>When does a foundation model used without any training beat a model fitted on the series? What I observed over a year of French electricity load, with interval coverage and not only the mean error, then rebuilding a router between the two on 50 series.</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://gwils28.github.io/posts/tsfm-zero-shot-specialiste/router_features.webp"/></item><item><title>A coding agent on a VPS: isolation, secrets and costs</title><link>https://gwils28.github.io/en/posts/agents-code-vps/</link><pubDate>Tue, 03 Nov 2026 00:00:00 +0000</pubDate><guid>https://gwils28.github.io/en/posts/agents-code-vps/</guid><description>A CPU VPS dedicated to the agent, reachable only over a private network, a separate production machine and GPUs rented by the hour: the state of play on 31 October 2026, from isolation to secrets and costs.</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://gwils28.github.io/posts/agents-code-vps/cover.png"/></item><item><title>Claude Code for data science: what changes in my working method</title><link>https://gwils28.github.io/en/posts/claude-code-methode-revisee/</link><pubDate>Thu, 29 Oct 2026 00:00:00 +0000</pubDate><guid>https://gwils28.github.io/en/posts/claude-code-methode-revisee/</guid><description>The lever is no longer the prompt but the context: what goes in, when, and what is handed to mechanisms that do not depend on the model. My method, revised as of 29 October 2026.</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://gwils28.github.io/posts/claude-code-methode-revisee/cover.png"/></item><item><title>Causal impact on time series: the counterfactual and its placebos</title><link>https://gwils28.github.io/en/posts/impact-causal-series-temporelles/</link><pubDate>Tue, 27 Oct 2026 00:00:00 +0000</pubDate><guid>https://gwils28.github.io/en/posts/impact-causal-series-temporelles/</guid><description>A counterfactual is a forecast. What I observed building one by hand, measuring its coverage on 100 panels with a known truth, planting four violations of the design, and applying it to Volkswagen&amp;rsquo;s share price after dieselgate.</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://gwils28.github.io/posts/impact-causal-series-temporelles/cover.png"/></item><item><title>Retraining frequency in forecasting: accuracy, freshness and stability</title><link>https://gwils28.github.io/en/posts/frequence-reentrainement/</link><pubDate>Sat, 17 Oct 2026 00:00:00 +0000</pubDate><guid>https://gwils28.github.io/en/posts/frequence-reentrainement/</guid><description>Accuracy, freshness and stability don&amp;rsquo;t call for the same retraining frequency. What I observed on a ten-policy bench, an analytical model, the Monash electricity data and a causal simulation.</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://gwils28.github.io/posts/frequence-reentrainement/cover.png"/></item><item><title>Hierarchical forecast reconciliation: promises and limits</title><link>https://gwils28.github.io/en/posts/reconciliation-hierarchique/</link><pubDate>Tue, 06 Oct 2026 00:00:00 +0000</pubDate><guid>https://gwils28.github.io/en/posts/reconciliation-hierarchique/</guid><description>MinT guarantees coherence, not accuracy, and not calibration. Shown through theory, simulation, French grid data and count sales.</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://gwils28.github.io/posts/reconciliation-hierarchique/cover.png"/></item><item><title>Claude Code for data science and ML-AI Engineering: my working method</title><link>https://gwils28.github.io/en/posts/claude-code-data-science/</link><pubDate>Tue, 29 Sep 2026 00:00:00 +0000</pubDate><guid>https://gwils28.github.io/en/posts/claude-code-data-science/</guid><description>A fast-coding agent doesn&amp;rsquo;t replace method. How I work with Claude Code: context as the scarce resource, verification as the one non-negotiable step.</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://gwils28.github.io/posts/claude-code-data-science/cover.png"/></item></channel></rss>