01
Retrieval that stays grounded
Uploaded PDFs are embedded and stored in Pinecone, so the assistant answers from the actual document via vector search rather than guessing.
← All case studies
GPT Glow
A PDF-based AI assistant with vector storage in Pinecone, OpenAI integration and Stripe-powered subscriptions — upload a document, then chat with it.
GPT Glow visual
Overview
GPT Glow turns any PDF into a conversation. I built the full SaaS — document upload, vector storage in Pinecone, OpenAI-powered retrieval and answers, persistent chat history, and Stripe subscriptions to monetize it — a complete RAG product end to end.
The Challenge
What made it hard.
Letting users "chat with a PDF" sounds simple but isn't: documents have to be chunked, embedded and retrieved accurately so answers are grounded in the source, not hallucinated.
And it had to be a business, not a demo — which meant subscription billing, persistent history and an interface anyone could use.
Approach
How I solved it.
02
Conversation with memory
Persistent chat history lets users return to a document and pick up where they left off — the assistant remembers the thread.
03
A real subscription product
Stripe-powered subscriptions and account management make GPT Glow a monetized SaaS, not just a clever demo.
What I Built
Shipped in the box.
- ✓PDF upload and document ingestion pipeline
- ✓Vector storage & semantic search via Pinecone
- ✓OpenAI-powered, source-grounded answers
- ✓Persistent per-document chat history
- ✓Stripe subscriptions and billing
Outcome
Where it landed.
RAG
End-to-end pipeline
Stripe
Subscription billing
Vector
Pinecone search
Have something like this to build?
I build AI-native SaaS, healthcare products, marketplaces and internal platforms from first commit to launch.