adityakarnam.com

adityakarnam.com

Source for adityakarnam.com — Aditya Karnam's personal site and running lab notebook. It's built less like a portfolio and more like a "world model": a set of linked pages (/systems, /stack, /field-notes, /now, /about) that track what's being built, the research questions behind it, and how it's evolving over time.

---

What's on the site

• Systems — public projects framed as evidence: research question, what was built, why it matters, with source links, not just a list of repos.
• Stack and Field Notes — the current research agenda and in-progress investigations, kept current rather than archived once-and-done.
• Now / About — current focus and background.
• Hero chat — a retrieval-augmented chat assistant on the homepage (functions/api/hero-chat.ts) that answers questions about the work using the site's own content as grounding, with persona support and multi-turn conversation.
• AI Toolkit — small standalone tools: an Intelligent Prompt Composer, a Prompt Grader & Rewriter, and a Tweet Thread Generator.
• Portfolio MCP — a hosted, read-only Model Context Protocol server (/mcp) that lets MCP clients query the site's projects and recent work as structured data.
• Blog — technical writing on AI/ML, agentic systems, Rust, and distributed systems, syndicated to DZone, Medium, and Level Up Coding.

Architecture

• Gatsby 5 + TypeScript on top of @lekoarts/gatsby-theme-minimal-blog, with the theme's components shadowed under src/@lekoarts/ where custom behavior is needed.
• Content lives in content/posts (MDX blog posts) and content/pages; structured "world model" data (systems, research agenda, field notes) lives in src/components/world-model/data.ts as typed, versionable data rather than freeform MDX.
• Deployed on Cloudflare Pages, git-integrated — every push builds and deploys automatically. Server-side behavior (hero chat, RAG retrieval) runs as Cloudflare Pages Functions under functions/.
• An AI Search / RAG corpus is synced from site content (systems, project pages, posts) to Cloudflare's managed AI Search via scripts/sync-rag-corpus.mjs, which backs the hero chat assistant with retrieval over the site's own public content.

Research

• ERBGA (Efficient Reduced-Bias Genetic Algorithm) — a novel genetic algorithm for generic community detection objectives; winner of a Master's research prize at UMSL.
• Published as "Efficient Reduced-Bias Genetic Algorithm (ERBGA) for Generic Community Detection Objectives."

Local development

npm install
npm run develop  # dev server at localhost:8000
npm run build    # production build
npm run serve    # serve the production build locally
npm test         # vitest

Other scripts:

npm run clean            # clear the Gatsby cache
npm run sync:rag-corpus  # manually trigger a RAG corpus sync (requires Cloudflare credentials)

Connect

• LinkedIn
• X (Twitter)
• GitHub
• Google Scholar

---

© 2026 Aditya Karnam