<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Thoughts on Machine Learning - Medium]]></title>
        <description><![CDATA[Explore the world of AI with ‘Thoughts on Machine Learning’: a high-quality blog offering in-depth tutorials and insights on AI startups. Ditch the hit-or-miss content of Medium for expert, well-researched articles that enlighten and inspire. - Medium]]></description>
        <link>https://medium.com/thoughts-on-machine-learning?source=rss----262f42e0ff64---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Thoughts on Machine Learning - Medium</title>
            <link>https://medium.com/thoughts-on-machine-learning?source=rss----262f42e0ff64---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 18:17:01 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/thoughts-on-machine-learning" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[AI agent evaluation shouldn’t require a PhD in infrastructure.]]></title>
            <link>https://medium.com/thoughts-on-machine-learning/ai-agent-evaluation-shouldnt-require-a-phd-in-infrastructure-9a6b5aac820e?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/9a6b5aac820e</guid>
            <category><![CDATA[ai-agents-in-action]]></category>
            <category><![CDATA[llm-evaluation]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[benchmark]]></category>
            <category><![CDATA[open-source]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Mon, 09 Feb 2026 23:34:42 GMT</pubDate>
            <atom:updated>2026-02-09T23:34:40.884Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mhezDiPYQC7omMPAeBRvmg.jpeg" /></figure><p>Running agent benchmarks is notoriously hard. Some need Docker containers. Some run for hours. Most harnesses are opaque, require dozens of dependencies, and assume you already know what you’re doing. This makes it incredibly difficult for newcomers to evaluate their agents properly.</p><p>That’s why I’m building <strong>Agent Harness</strong> — an open-source, local-first evaluation framework designed to be as simple as possible.</p><p>The first version is live:</p><pre>pip install agent-eval-harness</pre><p>What it does today:</p><ul><li>Write agents in <strong>any language</strong> (Python, JS, Rust, Go, Bash — anything with stdin/stdout)</li><li>Built-in benchmarks: <strong>Arithmetic, GAIA, Terminal-Bench</strong> (with Docker container grading)</li><li>Layered grading: exact → normalized → numeric → fuzzy → LLM-as-judge</li><li>Parallel execution with automatic retry</li><li>All logs are plain JSONL files — no external platform needed</li><li>One command to run: harness run --agent ./my_agent.py --benchmark gaia --model openrouter/deepseek/deepseek-chat-v3-0324</li></ul><p>Coming next:</p><ul><li><strong>AssistantBench</strong>, <strong>HLE</strong>, <strong>ARC-AGI</strong>, <strong>BrowseComp</strong> (already registered, wiring dataset loaders)</li><li>More execution modes: code submission, conversational, tool-use, GUI agents</li><li><strong>SWE-bench</strong>, <strong>GPQA</strong>, and more</li></ul><p>The goal is to <strong>democratize AI agent evaluation</strong>. Whether you’re a researcher, a hobbyist, or a team shipping agents to production — you should be able to benchmark your work in minutes, not days.</p><p>Check the repo, test it, break it, contribute:<br><a href="https://github.com/fsndzomga/agent-harness">https://github.com/fsndzomga/agent-harness</a></p><p>PyPI: <a href="https://pypi.org/project/agent-eval-harness/">https://pypi.org/project/agent-eval-harness/</a><br>DOI: <a href="https://doi.org/10.5281/zenodo.18568843">https://doi.org/10.5281/zenodo.18568843</a></p><p>Cite:</p><pre>@Misc{agentharness,<br>  title        = {Agent Harness: A Local-First, Lightweight Harness for AI Agent Evaluations},<br>  author       = {Franck Ndzomga},<br>  howpublished = {\url{https://github.com/fsndzomga/agent-harness}},<br>  doi          = {10.5281/zenodo.18568843},<br>  year         = {2026}<br>}</pre><p>#AIAgents #OpenSource #LLM #Evaluation #Benchmarks #MachineLearning</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9a6b5aac820e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thoughts-on-machine-learning/ai-agent-evaluation-shouldnt-require-a-phd-in-infrastructure-9a6b5aac820e">AI agent evaluation shouldn’t require a PhD in infrastructure.</a> was originally published in <a href="https://medium.com/thoughts-on-machine-learning">Thoughts on Machine Learning</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building AI Agents Is Not Enough]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/thoughts-on-machine-learning/building-ai-agents-is-not-enough-bb9d8d8b6887?source=rss----262f42e0ff64---4"><img src="https://cdn-images-1.medium.com/max/1448/1*IBrT4BCKPXQ47vj_wYCz2Q.png" width="1448"></a></p><p class="medium-feed-snippet">You need to do something simple, but easily ignored.</p><p class="medium-feed-link"><a href="https://medium.com/thoughts-on-machine-learning/building-ai-agents-is-not-enough-bb9d8d8b6887?source=rss----262f42e0ff64---4">Continue reading on Thoughts on Machine Learning »</a></p></div>]]></description>
            <link>https://medium.com/thoughts-on-machine-learning/building-ai-agents-is-not-enough-bb9d8d8b6887?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/bb9d8d8b6887</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[holistic]]></category>
            <category><![CDATA[princeton]]></category>
            <category><![CDATA[hal]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Wed, 09 Jul 2025 13:16:05 GMT</pubDate>
            <atom:updated>2025-07-09T13:16:05.766Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Constraints]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/thoughts-on-machine-learning/constraints-0a0c6978b939?source=rss----262f42e0ff64---4"><img src="https://cdn-images-1.medium.com/max/1024/1*SOyeNYHPDJfKpv3146uzng.png" width="1024"></a></p><p class="medium-feed-snippet">Sometimes in life, we stubbornly try to solve problems that can&#x2019;t be solved.</p><p class="medium-feed-link"><a href="https://medium.com/thoughts-on-machine-learning/constraints-0a0c6978b939?source=rss----262f42e0ff64---4">Continue reading on Thoughts on Machine Learning »</a></p></div>]]></description>
            <link>https://medium.com/thoughts-on-machine-learning/constraints-0a0c6978b939?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/0a0c6978b939</guid>
            <category><![CDATA[life]]></category>
            <category><![CDATA[mathematics]]></category>
            <category><![CDATA[problem-solving]]></category>
            <category><![CDATA[linear-algebra]]></category>
            <category><![CDATA[constraints]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Thu, 29 May 2025 13:38:55 GMT</pubDate>
            <atom:updated>2025-05-29T13:38:55.004Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[What Machine Learning Boils Down To]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/thoughts-on-machine-learning/what-machine-learning-boils-down-to-82591d52dbdc?source=rss----262f42e0ff64---4"><img src="https://cdn-images-1.medium.com/max/1536/1*304du5jMA5E0GkIRHHbvpA.png" width="1536"></a></p><p class="medium-feed-snippet">It&#x2019;s so simple you wouldn&#x2019;t believe it.</p><p class="medium-feed-link"><a href="https://medium.com/thoughts-on-machine-learning/what-machine-learning-boils-down-to-82591d52dbdc?source=rss----262f42e0ff64---4">Continue reading on Thoughts on Machine Learning »</a></p></div>]]></description>
            <link>https://medium.com/thoughts-on-machine-learning/what-machine-learning-boils-down-to-82591d52dbdc?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/82591d52dbdc</guid>
            <category><![CDATA[liberal-arts]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[agi]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Sat, 03 May 2025 21:49:15 GMT</pubDate>
            <atom:updated>2025-05-03T22:06:45.680Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Building an AI Financial Analyst Agent Using the smolagents Library]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/thoughts-on-machine-learning/building-an-ai-financial-analyst-agent-using-the-smolagents-library-100819097e52?source=rss----262f42e0ff64---4"><img src="https://cdn-images-1.medium.com/max/920/1*RT4AC2pxd1Lh0MdIQbW6Tw.png" width="920"></a></p><p class="medium-feed-snippet">It&#x2019;s so easy you&#x2019;ll be amazed.</p><p class="medium-feed-link"><a href="https://medium.com/thoughts-on-machine-learning/building-an-ai-financial-analyst-agent-using-the-smolagents-library-100819097e52?source=rss----262f42e0ff64---4">Continue reading on Thoughts on Machine Learning »</a></p></div>]]></description>
            <link>https://medium.com/thoughts-on-machine-learning/building-an-ai-financial-analyst-agent-using-the-smolagents-library-100819097e52?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/100819097e52</guid>
            <category><![CDATA[hugging-face]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[gpt-4o]]></category>
            <category><![CDATA[smolagents]]></category>
            <category><![CDATA[financial-analysis]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Tue, 29 Apr 2025 20:30:41 GMT</pubDate>
            <atom:updated>2025-04-29T20:30:41.149Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Llama 4 and Benchmark Lies]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/thoughts-on-machine-learning/llama-4-and-benchmark-lies-85f4445fac88?source=rss----262f42e0ff64---4"><img src="https://cdn-images-1.medium.com/max/2600/0*VeJT8xcEkj63rE0V" width="3573"></a></p><p class="medium-feed-snippet">AI Progress Is Overrated</p><p class="medium-feed-link"><a href="https://medium.com/thoughts-on-machine-learning/llama-4-and-benchmark-lies-85f4445fac88?source=rss----262f42e0ff64---4">Continue reading on Thoughts on Machine Learning »</a></p></div>]]></description>
            <link>https://medium.com/thoughts-on-machine-learning/llama-4-and-benchmark-lies-85f4445fac88?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/85f4445fac88</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[llama-4]]></category>
            <category><![CDATA[meta]]></category>
            <category><![CDATA[meta-ai]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Tue, 08 Apr 2025 15:28:00 GMT</pubDate>
            <atom:updated>2025-04-08T15:28:00.146Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Llama 4 and the Differentiation Game]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/thoughts-on-machine-learning/llama-4-and-the-differentiation-game-e21aeae59b7c?source=rss----262f42e0ff64---4"><img src="https://cdn-images-1.medium.com/max/1128/1*9WuuluB2HKjkNExfB_Kh6Q.png" width="1128"></a></p><p class="medium-feed-snippet">Leveraging Long Context and Cost Optimizations</p><p class="medium-feed-link"><a href="https://medium.com/thoughts-on-machine-learning/llama-4-and-the-differentiation-game-e21aeae59b7c?source=rss----262f42e0ff64---4">Continue reading on Thoughts on Machine Learning »</a></p></div>]]></description>
            <link>https://medium.com/thoughts-on-machine-learning/llama-4-and-the-differentiation-game-e21aeae59b7c?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/e21aeae59b7c</guid>
            <category><![CDATA[meta-ai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[meta]]></category>
            <category><![CDATA[llama-4]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Sun, 06 Apr 2025 01:05:24 GMT</pubDate>
            <atom:updated>2025-04-06T01:05:24.012Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Build a Chatbot with Internet Access using LangGraph]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/thoughts-on-machine-learning/build-a-chatbot-with-internet-access-using-langgraph-463a5c3e73c5?source=rss----262f42e0ff64---4"><img src="https://cdn-images-1.medium.com/max/2600/0*bar5515FcjOt93Xr" width="5724"></a></p><p class="medium-feed-snippet">Create an Intelligent Chatbot with OpenAI&#x2019;s GPT-4 and DuckDuckGo Search Integration</p><p class="medium-feed-link"><a href="https://medium.com/thoughts-on-machine-learning/build-a-chatbot-with-internet-access-using-langgraph-463a5c3e73c5?source=rss----262f42e0ff64---4">Continue reading on Thoughts on Machine Learning »</a></p></div>]]></description>
            <link>https://medium.com/thoughts-on-machine-learning/build-a-chatbot-with-internet-access-using-langgraph-463a5c3e73c5?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/463a5c3e73c5</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm-applications]]></category>
            <category><![CDATA[duckduckgo]]></category>
            <category><![CDATA[langgraph]]></category>
            <category><![CDATA[langgraph-tutorial]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Sun, 23 Mar 2025 14:31:21 GMT</pubDate>
            <atom:updated>2025-03-23T14:33:27.423Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Building a Columnar File Format from Scratch in Python (2)]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/thoughts-on-machine-learning/building-a-columnar-file-format-from-scratch-in-python-2-ab7523ae0f7f?source=rss----262f42e0ff64---4"><img src="https://cdn-images-1.medium.com/max/1312/1*0sgcByhkmKOS37HDjXAGXQ.png" width="1312"></a></p><p class="medium-feed-snippet">A Hands-On Approach to Better Understanding Data Warehouses</p><p class="medium-feed-link"><a href="https://medium.com/thoughts-on-machine-learning/building-a-columnar-file-format-from-scratch-in-python-2-ab7523ae0f7f?source=rss----262f42e0ff64---4">Continue reading on Thoughts on Machine Learning »</a></p></div>]]></description>
            <link>https://medium.com/thoughts-on-machine-learning/building-a-columnar-file-format-from-scratch-in-python-2-ab7523ae0f7f?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/ab7523ae0f7f</guid>
            <category><![CDATA[bigquery]]></category>
            <category><![CDATA[snowflake]]></category>
            <category><![CDATA[data-warehouse]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[data]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Wed, 19 Mar 2025 18:07:07 GMT</pubDate>
            <atom:updated>2025-03-19T18:07:07.337Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Vibe Coding Is Overrated]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/thoughts-on-machine-learning/vibe-coding-is-overrated-2aa0ed81e9cb?source=rss----262f42e0ff64---4"><img src="https://cdn-images-1.medium.com/max/1332/1*-YVvIpaMjHXArqnWcVE4pA.png" width="1332"></a></p><p class="medium-feed-snippet">Vibes Can Only Take You So Far</p><p class="medium-feed-link"><a href="https://medium.com/thoughts-on-machine-learning/vibe-coding-is-overrated-2aa0ed81e9cb?source=rss----262f42e0ff64---4">Continue reading on Thoughts on Machine Learning »</a></p></div>]]></description>
            <link>https://medium.com/thoughts-on-machine-learning/vibe-coding-is-overrated-2aa0ed81e9cb?source=rss----262f42e0ff64---4</link>
            <guid isPermaLink="false">https://medium.com/p/2aa0ed81e9cb</guid>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[code-quality]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[andrej-karpathy]]></category>
            <category><![CDATA[code]]></category>
            <dc:creator><![CDATA[FS Ndzomga]]></dc:creator>
            <pubDate>Tue, 18 Mar 2025 17:49:14 GMT</pubDate>
            <atom:updated>2025-03-18T17:52:12.661Z</atom:updated>
        </item>
    </channel>
</rss>