Vikas Udasi

AI & Cloud Architecture Β· Engineering Leadership Β· Open-Source AI Agent Builder @ Nielsen Marketing Cloud

prof_pic.jpg

πŸ“ Gurgaon, India

⚑ Building deep AI agents & cloud systems

πŸ”­ Open-source at github.com/vikasudasi

I’m an AI & Cloud Architecture leader at Nielsen Marketing Cloud, where I lead teams building scalable data systems, AI-driven pipelines, and the next generation of intelligent agents.

⚑ 20 open-source tools shipped β€” one every 2-3 days. I build open-source AI agent frameworks β€” tools that let developers run deep agents in their terminal, coordinate multi-agent workflows, and detect AI-generated content. Recent projects include:

  • πŸ—„οΈ skill-vault β€” Self-hostable, semantic skill registry for AI agents over a single MCP endpoint. Search, retrieve, and verify skills on demand.
  • πŸ–₯️ mcp-apps-render β€” CLI that renders MCP Apps interactive UI payloads (dashboards, forms, visualizations) directly in the terminal β€” no browser, no iframe, no HTML.
  • πŸ” slop-detect β€” Privacy-preserving CLI that detects AI-written text using 5 local heuristics. Fully offline, no API calls.
  • πŸ“Š prompt-riser β€” CLI that runs prompts through multiple reasoning strategies and visualizes the reasoning paths as a tree.
  • πŸ”¬ pr-classify β€” Detects AI-generated PRs and generates structured audit reports for human reviewers.
  • 🧹 skill-audit β€” Audits, validates, and inventories agent SKILL.md files across any ecosystem.
  • πŸ“Έ prompt-snapshot β€” Zero-dependency CLI that snapshots project context for AI agent prompts.
  • 🧠 my-agent β€” Personal macOS deep agent that lives in your terminal. Shell tools, MCP, voice, and subagent delegation.
  • πŸ“‹ AI Task Management System β€” Multi-agent coordination framework with CLI, web dashboard, and MCP server.

I’m deeply interested in multi-agent systems, MCP (Model Context Protocol), agent-native tooling, and the shift from brute-force AI hardware to algorithmic efficiency.

πŸ› οΈ Stack: Python Β· CLI (argparse/rich) Β· Jekyll Β· MCP Protocol Β· SQLite Β· Chromium CDP Β· GitHub Actions Β· LLM inference

When I’m not architecting cloud systems or building agents, I’m racing FPV drones, exploring new tech, and contributing to the open-source community.


Let’s build something ambitious πŸš€

I’m always open to discussing AI agents, cloud architecture, or open-source collaboration.

πŸ“§ vikasudasi@gmail.com Β· πŸ’Ό LinkedIn Β· πŸ™ GitHub

news

Aug 05, 2026 πŸ—„οΈ Launched skill-vault β€” self-hostable, semantic skill registry for AI agents over a single MCP endpoint. Search skills by task, retrieve what you need on demand, keep context small, and verify what you trust via sha256 + ed25519 signing. 111 tests, 88% coverage. GitHub Β· Read more
Aug 03, 2026 πŸ–₯️ Launched mcp-apps-render β€” CLI that renders MCP Apps interactive UI payloads (dashboards, forms, visualizations) directly in the terminal: no browser, no iframe, no HTML. Companion to mcp-app-suite, covering the headless/SSH/CI side of the same payload format. 53 tests, 85% coverage. GitHub Β· Read more
Aug 01, 2026 Just shipped doc-inject-guard β€” a CLI that scans Word docs, PDFs, markdown, and HTML for embedded prompt injection payloads targeting AI agents. The Context Collapse AI worm proved malicious instructions in DOCX files can alter Copilot’s financial data and self-propagate. This tool acts as a document-level security gate with five detection modules, risk scoring, and CI integration. Read more.
Jul 30, 2026 πŸ” Launched eu-act-check β€” CLI compliance scanner for EU AI Act Article 50. Scans files for C2PA manifests, EXIF metadata, and AI disclosure text patterns. GitHub
Jul 29, 2026 πŸ”§ Launched no-slop β€” CLI that scans your codebase for AI slop patterns (boilerplate comments, generic variables, over-abstraction, CSS spam) and generates constraint files for Claude Code, Cursor, Codex CLI, and Gemini CLI. Inspired by Hallmark, generalized across all frameworks. GitHub

latest posts