<?xml version='1.0' encoding='UTF-8'?><rss version='2.0'><channel><title>Vikram — Notes</title><link>https://vikrambtech2025-png.github.io/vikrambtech2025-png/blog/</link><description>AI/ML engineering notes</description><item><title>Idealistic agents, real-world problems</title><link>https://vikrambtech2025-png.github.io/vikrambtech2025-png/blog/idealistic-agents-real-problems.html</link><guid>vikram-idealistic-agents-real-problems</guid><pubDate>2026-10-12</pubDate><description>I love building <strong>AI agents that are idealistic</strong> — agents aimed at problems that
actually matter, not benchmarks that impress reviewers for a weekend.</description></item><item><title>RAG that cites itself</title><link>https://vikrambtech2025-png.github.io/vikrambtech2025-png/blog/rag-that-cites-itself.html</link><guid>vikram-rag-that-cites-itself</guid><pubDate>2026-10-05</pubDate><description>Most RAG demos retrieve a chunk and hope the model stays honest. I wanted the opposite:
a pipeline where <strong>every claim carries a receipt</strong>.</description></item><item><title>Shipping real AI on a 6 GB GPU</title><link>https://vikrambtech2025-png.github.io/vikrambtech2025-png/blog/shipping-on-6gb.html</link><guid>vikram-shipping-on-6gb</guid><pubDate>2026-09-24</pubDate><description>Constraint is a feature. When you only have <strong>6 GB of VRAM</strong>, you stop reaching for
the biggest model and start engineering.</description></item></channel></rss>