Hi HN, I love reading non-fiction, but I kept catching myself skimming, zoning out, and forgetting most of what I’d “read”. To fix that I built ThinkTotem, a small web app that lets you upload a PDF/EPUB/article/YouTube link and then chat with the material until you actually understand it.

What it does

- Ingests PDFs, EPUBs, Word docs, plain-text URLs and even YouTube transcripts (≤ 4 MB for now)

- Maps key ideas, trims filler and surfaces the essential concepts automatically

- Runs an active-recall loop: Socratic questions, explain-it-back prompts, spaced-repetition style refreshers

- Tracks mastery so you can skip, revisit or move on

Why not just use ChatGPT/Claude?

- General LLM chat is open-ended: you have to decide what to ask next.

- ThinkTotem is purpose-built for reading: short conversational loops keep attention, progress is visible, and the questions are sequenced to test retention rather than entertain.

How it works under the hood

- Document is paraphrased, chunked, metadata, summaries and questions are created, and stored in Postgres

- Every chat turn includes in the context the most relevant part of the current chapter, the summary of the book and the last messages (no vectorisation of the content needed)

- A small policy model classifies each turn (summary vs. question vs. recall) so the chat stays focused.

- A spaced-repetition scheduler writes “due” concepts back into the queue and surfaces them at the right time.

- User voice is transformed into text and LLM generated text is transformed into speech with OpenAI TTS/STT models

- All LLM logic runs server-side (Next.js app); the front end is Next.js with React Server Components.

Pricing / openness

There’s a free tier (3 uploads, 1 h chat, 50 messages) plus paid plans that just scale limits-no feature gating.

Costs

- Ingestion (Gemini Flash): ~30 s per 300-page PDF, $0.05–$0.10.

- Conversation (OpenAI TTS): ≈$0.40 per user-hour. Nothing is optimised yet; these two items dominate my bill.

Privacy

I delete originals after processing, just derivative work of it gets stored (e.g. summaries). I never feed your documents into model training and you can purge processed content at any time.

What’s missing / known issues

- 4 MB per-file cap-larger uploads are on the roadmap

- No org accounts yet; “Ultra” plan lets teams share a single login until then

- Conversation UX on mobile is still not state of the art (mobile Safari disables audio/mic when not in use for a few seconds, so UX is limited)

Ask

- Does the active-recall flow feel helpful or annoying?

- What would you need in the product to become a daily user?

Live demo

The app is live at https://thinktotem.com (no credit card needed).

I’ll be online all day-happy to answer anything, share infra costs (TTS is crazy expensive), or dive into the ingestion process.

Thanks for reading, -Claudio (solo maker, UTC+1)