<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Harshveer Singh: writing &amp; appearances</title><description>Director of Engineering at Elucidata, building AI systems for biomedical data.</description><link>https://harshveer.me/</link><language>en</language><atom:link href="https://harshveer.me/rss.xml" rel="self" type="application/rss+xml"/><item><title>AI as an investigator, not just a builder</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7519652216115208192/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7519652216115208192/</guid><description>This week, PHP released fixes for two critical vulnerabilities. The underlying issue had been suspected as early as 2024, but no one could show a concrete attack. Later AI-assisted research helped demonstrate how it could allow an attacker to execute code remotely, making the full impact clear.</description><pubDate>Sat, 24 Oct 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>Build AI capability inside the existing team</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7516367617264111616/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7516367617264111616/</guid><description>Sometimes the skills needed to build a product can change faster than we can organise the team building it. For engineering leaders, that creates a recurring decision: when the work demands a new capability, do we build a separate team around it, or develop it within the existing team?</description><pubDate>Thu, 15 Oct 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>Decision models make workflow gates affordable</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7513508929725898752/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7513508929725898752/</guid><description>Most agentic workflows run every step through a generative model. Even the steps that aren&apos;t generating anything. If we look closely, many steps are really decisions. Is this true or false? Which of these five categories does it fall in? How relevant is it, on a scale of one to five?</description><pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>Leave capacity uncommitted for innovation</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7509123718317658112/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7509123718317658112/</guid><description>One principle I generally push for in engineering planning is to leave some capacity uncommitted and leave room for discovery, experimentation, engineering growth. An engineering org that wants to be driven by innovation needs some capacity that is not already committed to a known deliverable.</description><pubDate>Fri, 25 Sep 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>AI changes the economics of legacy migrations</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7506941799668842496/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7506941799668842496/</guid><description>A lot of legacy systems stay in production because the cost of moving away from them is difficult to justify. The team knows where the limitations are, and often knows what a better stack would look like.</description><pubDate>Sat, 19 Sep 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>How our cloud bill kept falling while usage grew</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7503062764119298049/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7503062764119298049/</guid><description>At Elucidata our cloud bill has been going down for several quarters now, while usage and the product portfolio have been growing. When people ask what we did, they expect a technical answer, but most of it wasn&apos;t.</description><pubDate>Tue, 08 Sep 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>Let AI agents control what they reveal, not what they read</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7501278196714209280/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7501278196714209280/</guid><description>Should we let AI Agents censor their own output instead of applying access control filters at source? Idea is uncomfortable, but I think we will end up there anyway. Most companies now have an AI agent over their internal knowledge base, answering questions from whatever has been indexed into it.</description><pubDate>Thu, 03 Sep 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>Sending AI-generated text unedited is a cost transfer</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7500087771919908864/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7500087771919908864/</guid><description>A very common use of AI now is generating explanations, summaries, design reasoning, brainstorming output. And by default, most reasoning models return several paragraphs even when the actual answer is a line or two. The model doesn&apos;t know what you already know.</description><pubDate>Mon, 31 Aug 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>Versioned knowledge graphs: why an agent&apos;s shortlist changes</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7495498229162303488/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7495498229162303488/</guid><description>Ask an agent to shortlist targets for an indication today. Ask again in 6 months and two names have dropped off. Nothing about the question changed. What changed — the biology, the source databases, or the ingestion underneath? On most systems nobody can say.</description><pubDate>Tue, 18 Aug 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>MCP servers for Polly</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7475064494637989888/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7475064494637989888/</guid><description>For years, we&apos;ve delivered AI-ready biomedical data that teams trust to drive discovery. Now we&apos;re handing them the controls — letting people build the exact workflows their science demands.</description><pubDate>Tue, 23 Jun 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>Security blind spots in enterprise RAG</title><link>https://www.linkedin.com/feed/update/urn:li:activity:7447852190242611200/</link><guid isPermaLink="true">https://www.linkedin.com/feed/update/urn:li:activity:7447852190242611200/</guid><description>Most enterprise RAG deployments have security blind spots that should keep you up at night.</description><pubDate>Thu, 09 Apr 2026 00:00:00 GMT</pubDate><category>LinkedIn</category></item><item><title>Federated Learning in Healthcare AI</title><link>https://www.elucidata.io/whitepapers/federated-learning-healthcare-ai</link><guid isPermaLink="true">https://www.elucidata.io/whitepapers/federated-learning-healthcare-ai</guid><description>Learn how Elucidata’s federated learning enables secure, compliant AI-ready data collaboration - faster insights, lower costs and better generalizability.</description><pubDate>Mon, 08 Sep 2025 00:00:00 GMT</pubDate><category>Elucidata whitepaper</category></item><item><title>Scaling High-Quality Data Processing: How to Achieve 4x Cost Reduction for Foundation Models</title><link>https://www.elucidata.io/webinars/scaling-compute-for-building-foundation-models-in-biology</link><guid isPermaLink="true">https://www.elucidata.io/webinars/scaling-compute-for-building-foundation-models-in-biology</guid><description>This presentation will demonstrate how Polly achieves 4x lower costs and 2.5x faster processing speeds. Gain insights into a head-to-head comparison of Polly and AWS on cost, speed, and ease of use.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate><category>Elucidata webinar</category></item><item><title>Security &amp; Compliance on Polly - Accelerating Drug Discovery Securely</title><link>https://www.elucidata.io/whitepapers/security-compliance-on-polly-accelerating-drug-discovery-securely</link><guid isPermaLink="true">https://www.elucidata.io/whitepapers/security-compliance-on-polly-accelerating-drug-discovery-securely</guid><description>Explore how Polly ensures security and compliance for accelerating drug discovery with robust data management.</description><pubDate>Mon, 12 Aug 2024 00:00:00 GMT</pubDate><category>Elucidata whitepaper</category></item><item><title>Role of MLOps in Biomedical Research</title><link>https://www.elucidata.io/blog/role-of-mlops-in-advancing-biomedical-research</link><guid isPermaLink="true">https://www.elucidata.io/blog/role-of-mlops-in-advancing-biomedical-research</guid><description>Discover how MLOps is helping advance biomedical research, enhance drug discovery, and improve healthcare outcomes.</description><pubDate>Wed, 07 Aug 2024 00:00:00 GMT</pubDate><category>Elucidata article</category></item><item><title>What is the right size of a team?</title><link>https://harshveer.me/writing/right-team-size/</link><guid isPermaLink="true">https://harshveer.me/writing/right-team-size/</guid><description>I have faced this question multiple times whenever there is a change in team or new team is being created. I am sure other engineering managers(at junior level or senior level) face this situation.</description><pubDate>Mon, 01 Jul 2024 00:00:00 GMT</pubDate><category>Essay</category></item><item><title>Changing User Account Password in Ubuntu</title><link>https://harshveer.me/writing/changing-user-account-password-in-ubuntu/</link><guid isPermaLink="true">https://harshveer.me/writing/changing-user-account-password-in-ubuntu/</guid><description>Changing your user account password in Ubuntu is fairly simple...</description><pubDate>Tue, 14 Nov 2023 00:00:00 GMT</pubDate><category>Essay · Ubuntu</category><category>archive</category><category>ubuntu</category></item><item><title>Embedding Terminal in Gedit</title><link>https://harshveer.me/writing/embedding-terminal-in-gedit/</link><guid isPermaLink="true">https://harshveer.me/writing/embedding-terminal-in-gedit/</guid><description>You can embed terminal right in gedit in it&apos;s bottom panel.</description><pubDate>Wed, 01 Feb 2023 00:00:00 GMT</pubDate><category>Essay · Ubuntu</category><category>archive</category><category>ubuntu</category></item><item><title>Automatically Mounting Partitions in Ubuntu</title><link>https://harshveer.me/writing/automatically-mounting-partitions-in-ubuntu/</link><guid isPermaLink="true">https://harshveer.me/writing/automatically-mounting-partitions-in-ubuntu/</guid><description>If you have many partitions of your disk then you can see that when you start Ubuntu all partitions doesn&apos;t get mounted automatically.</description><pubDate>Thu, 03 Mar 2022 00:00:00 GMT</pubDate><category>Essay · Ubuntu</category><category>archive</category><category>ubuntu</category></item></channel></rss>