DocentAi-1
Team GluconD
December 29, 2025
Subtitle: A look at how Kiro helped me architect my RAG application before writing a single line of code.
Introduction
We’ve all been there: you start a hackathon project with create-next-app, code furiously for 4 hours, and then realize your data structure is completely wrong. For my latest project, DocentAI (a RAG tool that lets developers chat with live documentation), I decided to try a different approach. Instead of rushing into code, I used Kiro to build a “Blueprint” first. Here is how focusing on Architecture-First saved me hours of debugging.
Phase 1: The Blueprint (Powered by Kiro)
Before I installed a single npm package, I used Kiro to map out my entire project structure. Kiro isn’t just a note-taking app; it lives right inside my workflow, helping me link my “thinking” to my “coding.” I created a roadmap.md and an architecture.md file to visualize how the
Caption: My architectural plan in Kiro, living right next to the source code
By doing this, I caught a major issue early: I realized I needed a robust client-side storage solution for chat history. This led me to add IndexedDB to the stack before I hit a wall during development.
Phase 2: The Tech Stack Once the plan was solid, the setup was seamless. Here is what I’m using: 1 • Framework: Next.js 14 (App Router) • Styling: Tailwind CSS + Framer Motion (for those smooth sidebar ani- mations) • State Management: Jotai (Atomic state is perfect for the sidebar logic) • Database: PostgreSQL + Prisma • Planning: Kiro
Phase 3: Building the “Skeleton” I’ve just completed the core interface. The goal was to build a distraction-free environment where the “Chat” is the hero. I spent time refining the Sidebar logic to handle history states without re- rendering the whole page. Using framer-motion, I added smooth transitions when opening/closing the menu.
What’s Next? Now that the frontend infrastructure is live (including Authentication and Database helpers), I’m moving on to the “Brain” of the operation: the RAG integration. I’ll be connecting the AI response logic and setting up the vector embeddings next.
Check out the progress on GitHub: https://github.com/tushar-404/DocentAI