https://dipankarnath.pages.dev/tags/llm/Recent content in Llm on Dipankar NathHugoen-usMon, 18 May 2026 00:00:00 +0000https://dipankarnath.pages.dev/posts/codewiki-bot/Mon, 18 May 2026 00:00:00 +0000https://dipankarnath.pages.dev/posts/codewiki-bot/<p><strong>CodeWiki Bot</strong> is an AI-powered documentation and chat assistant that sits on top of a codebase. Instead of hunting through repos and outdated wikis, developers just <em>ask</em> — and get grounded answers, cross-repo dataflow diagrams, and architecture context on demand.</p> <p><img alt="CodeWiki Bot dashboard" loading="lazy" src="https://dipankarnath.pages.dev/posts/codewiki-bot/codewiki-bot.png"></p> <h2 id="what-it-does">What it does</h2> <ul> <li><strong>Graph-RAG engine</strong> — combines a knowledge graph of the code with retrieval-augmented generation, so answers understand <em>relationships</em> between modules, not just keyword matches.</li> <li><strong>Cross-repo dataflow &amp; architecture diagrams</strong> — automatically generated from indexed code, spanning multiple repositories.</li> <li><strong>Natural-language chat over the codebase</strong> — ask &ldquo;where is X handled?&rdquo; or &ldquo;how does data flow from A to B?&rdquo; and get a sourced answer.</li> <li><strong>Semantic search across Azure DevOps repos</strong> — 12,000+ documents indexed and kept in sync.</li> </ul> <h2 id="why-graph-rag-not-plain-rag">Why Graph-RAG (not plain RAG)</h2> <p>Plain RAG retrieves similar text chunks. A <strong>graph</strong> layer adds the <em>structure</em> of the codebase — call relationships, data flow, module dependencies — so the bot can reason across files and repos instead of answering from isolated snippets. That&rsquo;s what makes cross-repo diagrams and &ldquo;how does this connect&rdquo; questions possible.</p>