I have ADHD and a 200+ book TBR pile. The problem is I keep putting down complex novels, then picking them up three weeks later with absolutely no memory of who these characters are or why they're arguing about inheritance law.

So I built Where Was I? Upload a PDF or EPUB, tell it what page you're on, and it generates a spoiler-free summary of everything up to that point.

Live: https://app.wherewasi.co.uk (no signup required - just upload and go)

The system uses RAG with semantic chunking that follows narrative structure rather than arbitrary page breaks. It extracts plot points using MMR and contextual relevance scoring, then generates summaries that avoid spoiling anything beyond your current page.

Tech stack: - FastAPI backend on Railway, Google Cloud Storage for frontend - ChromaDB for vector storage with custom book collections - OpenAI embeddings (text-embedding-3-small) for semantic search - GPT-4.1 for summarisation that only sees pages you've read to prevent spoilers - Semantic chunking using semchunk (respects sentence boundaries, not just tokens)

This is genuinely v0.1. The UI is basic, it will probably crash under load, and I'm sure there are edge cases I haven't considered. But it works. I use it every time I pick up a book I abandoned.

Built this because I got tired of either (a) giving up on complex books or (b) spending 20 minutes flipping back through chapters trying to remember if the butler was suspicious or just socially awkward.

Happy to answer questions about the RAG implementation, spoiler prevention strategies, etc.