You are browsing as a guest. Sign up (or log in) to start making projects!

DEEPREAD

  • 10 Devlogs
  • 16 Total hours

Building DeepRead, an AI-powered research assistant that helps users understand long PDFs. Instead of just summarizing documents, it creates an interactive knowledge graph, lets users ask questions, and shows the exact evidence from the PDF for every answer.

Ship #2

DeepRead — upload a PDF and get a summary, flashcards, a quiz, a mind map, and a chat where you can ask questions about it.

Stopping the AI from making up facts that weren't in the PDF, and dealing with free AI services running out of quota (so I made it try 4 different AI providers one by one until one works).

That it actually works with real PDFs end to end, not just a demo — plus the flip flashcards and the fallback system feel solid.

Upload any PDF, try all 3 summary modes, flip through flashcards, take the quiz, check the mind map, and ask the chat something not in the PDF to see it say "not covered" instead of guessing.

  • 5 devlogs
  • 11h
Try project → See source code →
Open comments for this post

2h 45m 30s logged

DeepRead v2 is now finished. I’ve had a little bit of time to test it all out again, make sure there are no bugs and that all the features work well together. All in all, it has developed greatly from its initial stage and I feel quite proud about the result. There will definitely be minor improvements that can still be made but, for now, I consider it to be fully completed and ready for release.

0

Loading discussion…

0
16
Open comments for this post

3h 8m 45s logged

Worked on the services folder today and expanded the AI provider support for DeepRead v2. The app previously had three AI providers, and I started integrating another one to make the fallback system more reliable. I finished the implementation for two of the providers and tested them individually. There’s still a bit of integration left, but the services layer is coming together nicely and should make the app much more dependable.

0

Loading discussion…

0
6
Open comments for this post

43m 43s logged

I finished writing the code for each feature individually and made sure everything works on its own. The only major thing left now is integrating everything into the main application and testing how the features work together. Once that’s done, I’ll start polishing the overall experience and fixing any issues that come up during testing.

0

Loading discussion…

0
7
Open comments for this post

2h 45m 9s logged

Made some more progress on DeepRead v2 today. Most of the frontend is now in place and I finished the HTML, CSS and most of the JavaScript. I actually spent most of my time tracking down a really annoying JavaScript bug that was breaking part of the new flow. Took longer than I expected, but it finally works the way I wanted. The base for the new features is ready now, so the next step is polishing everything and connecting the remaining functionality.

0

Loading discussion…

0
9
Open comments for this post

2h 0m 18s logged

Today I started working on the frontend for DeepRead v2. I didn’t want to completely redesign the app, so I kept the same clean look while improving the overall layout and spacing. Most of the work today went into HTML and CSS, making everything feel more organized and easier to use. I also planned the layout for the upcoming features so adding them later will be much smoother. Still a lot left to build, but the base for v2 is ready now.

0

Loading discussion…

0
10
Ship #1

DeepRead — a tool that uploads a PDF and generates an AI summary and flashcards so you don't have to read the whole thing.
Gemini kept hitting quota limits so I had to rebuild the backend to support multiple AI providers. Then AI models returning broken JSON, flashcards showing old PDF data, CORS issues, frontend not updating — basically spent most of my time debugging.

The multi-AI fallback system. Groq fails → tries OpenRouter → tries Gemini. Happens automatically, user doesn't even notice. Also just the fact that I actually finished it.
You need all three API keys (Groq, OpenRouter, Gemini) in a .env file. Text-based PDFs work best, not scanned ones. Larger files take longer. Start the FastAPI backend first, then open index.html directly in your browser.

Try project → See source code →
Open comments for this post

28m 3s logged

Small Improvements

Spent some more time improving DeepRead today.I wasn’t adding any major features, but I cleaned up the project and fixed a few small issues that I noticed while testing.

I also improved the AI fallback logic a bit and organized some backend files to make the code easier to maintain.

I tested the upload, summary and flashcard flow a few more times to make sure everything works smoothly before wrapping up the project.
Nothing huge today, just polishing the project and making it more stable. There are still a few ideas I want to work on later, but I’m happy with where DeepRead is now

0

Loading discussion…

0
12
Open comments for this post

2h 36m 38s logged

Devlog #Final – DeepRead is Done (Finally)

I can’t believe I’m actually writing this. After weeks of “just one more bug” and “why is this not working”, DeepRead is actually complete. Like, functional complete. Not “it works on my machine” complete.

The idea was dead simple from the start — I just wanted something that reads my PDFs and gives me a summary so I don’t have to. That’s it. Didn’t think it would turn into this whole thing.

But man, the number of times this project almost broke me…

The biggest headache was the AI APIs. I started with Gemini because it was the easiest to set up. Worked great for like two days. Then the quota limits hit. Every. Single. Time. So I had to completely restructure the backend to support multiple providers. Now it tries Groq first, falls back to OpenRouter, and if both fail, Gemini is the last resort. Took way too long to get this right but honestly it’s the most solid part of the app now.

Then there was the JSON parsing nightmare. You’d think AI models would return clean JSON when you ask them to. They don’t. They really don’t. Sometimes there’d be extra text before the JSON, sometimes missing brackets, sometimes completely wrong format. Spent hours writing fallback parsers for that stuff.

And don’t even get me started on the frontend. There was this one bug where the flashcards would show data from a PREVIOUS PDF. Took me an entire evening to figure out I wasn’t clearing state properly. Fun times.

What’s actually in it right now:

  • PDF upload (shockingly, this was the easiest part)
  • AI summary generation
  • AI flashcards with prev/next navigation
  • Multiple AI fallback system (the thing I’m most proud of ngl)
  • Simple frontend — HTML, CSS, vanilla JS
  • FastAPI backend

No frameworks, no unnecessary complexity. Just works.

What I want to add later:

  • Dark mode (I keep procrastinating on this)
  • PDF preview so you don’t have to open it separately
  • Highlighting key sections
  • Quiz mode — flip the flashcards into actual questions
  • Export summary as PDF or Markdown
  • User accounts to save study history
  • DOCX and PPT support
  • Better loading states and error messages (currently they’re… basic)

This project started because I was lazy and didn’t want to read long PDFs. But it ended up teaching me more than most tutorials ever did. Real debugging, real API integration problems, real “why is CORS blocking this when it worked yesterday” energy.

If you’re thinking about building something like this — just start. You’ll hate it at points, but you’ll learn so much.

Anyway, that’s the devlog. DeepRead v1.0 is done. Time to actually use it for my studies now 😅

Thanks for following along. 🚀

0

Loading discussion…

0
4
Open comments for this post

25m 8s logged

Started reading uploaded PDFs

Today I added the first step of PDF processing.
After uploading a PDF, DeepRead now extracts the text and returns a preview.
This is the foundation for summaries, search, question answering, and the knowledge graph that I’ll build next.

0

Loading discussion…

0
4
Open comments for this post

23m 54s logged

Connected the frontend with the backend

Got the first end-to-end feature working today. Users can now select a PDF, upload it from the website, and the backend saves it successfully. It took a bit of debugging, but the upload flow is now working as expected.

Next is extracting text from the uploaded PDF so DeepRead can start understanding the document.

0

Loading discussion…

0
15
Open comments for this post

45m 39s logged

Started building DeepRead

Today I set up the project and got the basic structure ready

  • Created the GitHub repository
  • Set up the FastAPI backend
  • Created the frontend
  • Built the first homepage with a PDF upload section
  • Added the basic upload flow on the frontend

Next, I’ll connect the upload button to the backend and start processing PDF files.

0

Loading discussion…

0
19

Followers

Loading…