🧠 Promptly – Build powerful AI apps from your data. No code. Just launch.

🚀 Inspiration

We were driven by a simple question: What if anyone could create their own ChatGPT, trained on their documents, in minutes – no code required?
As AI becomes integral to knowledge work, we saw creators, teams, and small businesses struggle to build AI copilots without engineering help. Promptly was born to solve this — a no-code platform where you can build and launch document-trained bots effortlessly.

🧰 What it does

Promptly lets users:

• Upload documents (PDF, Word, etc.)
• Automatically parse and embed content using Pinecone
• Create AI bots using OpenAI that chat with the uploaded content
• Access and manage bots through a modern UI
• Sign up with email or anonymously to try the platform

Each bot gets a personalized introduction, memory of its document context, and a live chat interface — creating an intelligent, always-on assistant.

🏗️ How we built it

• Frontend: Built with Next.js 15, Tailwind CSS, and Radix UI for a responsive, accessible user interface.
• Authentication & DB: Handled by Supabase, with custom logic for email auth, guest sessions, and document storage.
• LLM Integration: We use OpenAI APIs for generating chat responses and Pinecone for vector search from ingested content.
• File Parsing: Leveraged Mammoth for DOCX and pdf-parse for PDFs, streaming into vector embeddings.
• Chat Logic: A custom AI message handler routes prompts and retrieval logic to create seamless, contextual replies.

🧱 Built With

• Languages: TypeScript
• Frameworks: Next.js, Tailwind CSS, React
• Cloud Services: Supabase (auth, storage, DB), Vercel (hosting), Pinecone (vector DB)
• APIs: OpenAI, Stripe (for upcoming billing)
• Packages: mammoth, pdf-parse, @supabase/supabase-js, @pinecone-database/pinecone, react-hook-form, zod

🚧 Challenges we ran into

• Handling secure server-side Supabase logic in a Next.js environment (avoiding admin key leaks)
• Creating dynamic chat context based on document structure and ensuring latency stayed low
• Integrating PDF and DOCX parsing cleanly across browser and server environments
• UI personalization for bots — how to make each one feel unique and aligned with its content

🌟 Accomplishments we're proud of

• Created a seamless document-to-chatbot flow with zero code
• Enabled real-time embeddings and context injection for long documents
• Built a guest-friendly experience that encourages experimentation
• Architected the platform for future monetization and analytics integration

📚 What we learned

• Supabase makes rapid development possible but requires discipline around admin/client context separation
• Tailoring LLM behavior through document context improves accuracy more than prompt engineering alone
• Developer UX is key — our next users are builders, not coders

🔮 What's next for Promptly

• Add custom domain support so users can embed bots on their sites
• Build organization-level workspaces for teams
• Introduce feedback and retraining loop to refine bot responses
• Add more AI models (Claude, Gemini) and agent-based features
• Launch billing, analytics, and versioning for enterprise users

Built With

• bolt
• css
• gemini
• langchain
• lucide
• mammoth
• next.js
• pdf-parse
• pinecone
• rag
• react
• recharts
• slack
• stripe
• supabase
• tailwind
• typescript
• vercel