Divyansh Shukla

1Selected work

Six products I built and shipped in 2026, each live or in public beta. Everything else, from open-source tools to client sites, follows in Table 1, and every project has a fuller entry on the work page.

(a) Ice · ice.naravirtual.aiBeta · $5 reservations

(b) Blank · nvblank.comLive · free to start

(c) Synapse · naravirtual.ai/synapseLive · free plan

(d) CampusQuiz · campusquiz.inLive · free for students

(e) CampusPrep.in · campusprep.inLive · free

(f) Saakshya · saakshya.divyanshshukla.comLive demo

2Research

Plan Once, Ground Locally: Plan-and-Execute Web Agents Match ReAct Accuracy at One-Sixth of the Token Cost. Divyansh Shukla. Preprint, Zenodo, 23 September 2026. doi:10.5281/zenodo.22904595.

LLM web agents usually follow the ReAct pattern: the model is called again after every browser action, so each click is paid for with a network round trip and with a prompt that has grown since the last one. We study the alternative in which the model writes the whole plan in one call and the steps are executed locally, with a small on-device decision model (Laya, a ModernBERT-large “System-1” model used zero-shot) mapping each step to a page element, judging when asynchronous content has arrived, and checking actions that left the page unchanged.

Against the budget-matched baseline, planning once is not significantly different on success (88.8% vs 92.0%; paired difference −3.2 points, 95% CI −7.7 to +1.2) while using 83.6% fewer input tokens, 67.3% fewer LLM calls and 80.4% less money, and finishing 4.1× faster.

@misc{shukla2026planonce,
  title     = {Plan Once, Ground Locally: Plan-and-Execute Web Agents
               Match ReAct Accuracy at One-Sixth of the Token Cost},
  author    = {Shukla, Divyansh},
  year      = {2026},
  month     = sep,
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22904595},
  url       = {https://doi.org/10.5281/zenodo.22904595},
  note      = {Preprint}
}

Findings and full abstractPDFCode and data

3Notes

Short working papers, with the numbers behind them. New ones arrive on the notes page and by RSS.

• What 2,800 browser-agent runs taught me about planning. 3 October 2026 Plan-and-execute web agents matched a budget-matched ReAct baseline on 200 tasks with 83.6% fewer input tokens. What held up, what didn’t, and what to copy.
• Making my site readable by agents. 3 October 2026 This site serves an MCP server, WebMCP tools, llms.txt and a markdown copy of every page, all from one content source. What each piece does and how to try it.

4Reading this site with an agent

Every page here has machine-readable twins, generated from the same source as the HTML, so the versions never disagree. There are three ways in.

MCP. A read-only server over Streamable HTTP at /mcp. To add it to Claude Code:

claude mcp add --transport http divyansh https://divyanshshukla.com/mcp

Cursor, VS Code and other clients take the URL https://divyanshshukla.com/mcp directly; stdio-only clients can bridge with mcp-remote. Table 3 lists the tools.

WebMCP. Agents running inside your browser get the same tools on this page, registered through WebMCP, with no setup.

Markdown and llms.txt. Append .md to any URL, or send Accept: text/markdown. /llms.txt maps the site and /llms-full.txt is all of it in one file.

curl -H "Accept: text/markdown" https://divyanshshukla.com/

5Background

B.Tech, Computer Science (AI & Data Science). Earlier projects, a voice assistant, a gesture-controlled mouse, wireless-security research and a bug-bounty toolkit, are in Table 1. The tools I reach for:

Languages

Python, TypeScript, JavaScript, SQL

Agents and LLMs

Anthropic and OpenAI APIs, OpenRouter, MCP, Playwright, LangChain, RAG with ChromaDB and Pinecone

Models on device

MLX, ModernBERT, MediaPipe, OpenCV

Web

SvelteKit, Next.js, FastAPI, Django, Tailwind CSS

Data and infrastructure

Postgres, Redis, Docker, Dokploy, Vercel, Cloudflare

Security and hardware

Aircrack-ng, Hashcat, ESP32, Raspberry Pi

6Correspondence

Open to AI engineering roles, and to conversations with investors and partners about these products. Based in India; remote works. I usually reply within a day.

[email protected]Copy

LinkedInGitHubAll workNotes