Inspiration
Bringing Dead Docs to Life We’ve all lived through the nightmare of being the "new hire." You join a company, and you're immediately buried under a graveyard of PDF handbooks, obsolete wikis, and scattered Markdown files. You feel like a zombie wandering the halls, afraid to pester the senior devs with "stupid questions." For Kiroween, I wanted to build a Frankenstein's Monster of a solution: stitching together the best modern web tech (Next.js) with powerful AI (Gemini) to resurrect those dead documents into a living, breathing, interactive guide. Thus, quickOnboardDoc was summoned.
What it does
quickOnboardDoc is a secure, multi-tenant RAG (Retrieval-Augmented Generation) chatbot.
Ingest: Admins upload company docs (PDF, TXT).
Process: The system chunks and embeds this data into a vector store.
Resurrect: New hires chat with the bot. The system performs vector similarity searches to find the exact context and uses Google's Gemini API to generate a grounded answer.
PWA (Mobile Companion): Crucially, this is built as a Progressive Web App (PWA). New hires can install it directly to their phone's home screen, giving them a pocket-sized mentor that works just like a native mobile app.
How I built it
I stitched together a chimera of technologies to make this work:
The Body (Frontend): Next.js (App Router) for the framework, styled with Tailwind CSS and shadcn/ui. I used Framer Motion to give it that "Crypt Keeper" violet glow and haunting entrance animations for the Costume Contest bonus.
The Brain (AI & Logic): I used LangChain for document orchestration and Google Gemini 2.5 Flash for the intelligence.
The Spine (Data & State): Prisma + PostgreSQL handles the multi-tenant relational data, while TanStack Query manages the complex async state, persisting it to session storage so the app feels instant.
The Soul (Kiro): This project was built solo using Kiro. I utilized a strict global_steering_rules.md file to force Kiro to adhere to my specific stack (NextAuth, Prisma, TanStack Query) without needing constant correction.
Challenges I ran into
The PWA Configuration: Making the app installable and responsive across devices while maintaining the heavy animations of the "Crypt Keeper" theme was tricky. I had to carefully tune the manifest and service workers to ensure it felt like a native mobile app.
Hallucinations: Early versions of the bot would invent company holidays. I fixed this by implementing strict "System Instructions" via the Gemini API, forcing it to stick strictly to the retrieved context.
Accomplishments that I'm proud of
PWA Certification: Successfully making the app installable. You can add it to your iOS or Android home screen right now!
Kiro Steering: I effectively "cloned" myself by writing a Kiro Steering Doc. Kiro started writing code exactly the way I do, down to the specific import order.
The Aesthetic: It's rare to see enterprise software that looks cool. I love the high-contrast violet "dark mode" theme.
What I learned
building an idea and an mvp can be super hard but with agentic development and thanks to kiro streed.md I was able to spring up a demo(mvp) in less than a day and make changes faster for improvements and solving complex bugs was far easier because I had access to very smart agents and tools to debug and fix and be able to ship out solutions faster.
What's next for quickonboarddoc
Voice Mode: Adding speech-to-text so you can talk to the "ghost" in the machine.
Slack Integration: Bringing the bot directly into team channels.
More File Types: Adding support for parsing Excel and CSV files for HR data.
Built With
- ai
- cssgoogle-gemini-api
- framer
- kiro-(ai-ide)
- langchain
- next.js
- prisma
- pwa
- react
- sdkpostgresql
- shadcn-ui
- tailwind
- typescript
- vercel
- web
Log in or sign up for Devpost to join the conversation.