Current Mission

const shuchita = Engineer(
  experience: '4 years Flutter in production',
  focus: [
    'offline-first architecture',
    'agent workflows for real delivery pipelines',
    'context engineering',
  ],
  mobileStack: ['Flutter', 'Dart', 'Firebase', 'SDUI', 'offline-first architecture'],
  aiStack: ['agent orchestration', 'CI-grade acceptance gates', 'MCP'],
  currentBuild: 'story-agent, planning core for AI-augmented delivery',
  writingAt: 'medium.com/@coderSJ',
  mode: 'planning-first, measurable outcomes, architecture before automation',
);

Flagship: Story Agent

story-agent is the anchor project. A planning agent that converts raw user stories into scoped, testable delivery tasks, built around the constraints that show up in real production teams: partial context, shifting specs, and humans who need to stay in the loop.

Project Constellation

Fallback docking index:

• Center ⭐ → awesome-ai-setup
• Top Left → story-agent
• Top Right → pr-scout
• Bottom Left → MovieDeck
• Bottom Right → tech-debt-agent

Writing

Your AI Isn't Getting Dumber. Your Session Is Getting Dirtier. AI Won't Replace Flutter Developers. The Devs Across The Table Will. Google's A2UI Is SDUI With an LLM. Mind the Scars. The SDK Tax: A Bill You Pay In Outages, Not Invoices.

Active Research

⚙️ Agent Infrastructure Agents forget. They drift. They confidently act on context that stopped being true three tool calls ago. What does it actually take to keep one coherent?	🛡️ Safety & Governance The scariest prompt injections don't look like attacks. They look like normal user input. Where do you even draw the line, and who enforces it?	��� Agent Observability If your agent makes a bad call at 2am, can you reconstruct exactly what it saw, what it decided, and what it cost? Most setups can't.
🧠 Codebase Intelligence A codebase isn't a document. It's a set of decisions made by people who've since left. Getting an agent to understand that, without reading everything, is the unsolved part.	🚀 AI-Augmented Delivery Adding an agent to a team is easy. Not breaking the team's existing trust and review culture while doing it, that's the part nobody writes about.	📱 Mobile × AI Mobile apps have hard constraints: offline states, small screens, slow networks. Most AI tooling ignores all of that. What changes when it can't?

GitHub Signals