this app runs entirely in your browser.
thanks to Codex, it now uses a YOLO detector using ONNX to detect regions in each rendered PDF page image (text, pictures, formulas, tables).
based on this analysis, PDF Reflow then cuts out TEXT regions into tiny little word images and adds back the original text from the PDF (when available) so you can still select and find text in the reflowed HTML render.
Hope you find this useful.
The original version used heuristics, but frankly it was breaking on many more PDF that I'd want to admint.
This new AI-based version that uses a YOLO detector trained specifically on Doc Layout dataset seems to do a very very good job.
Cooking a local AI version of PDF Reflow to show PDFs on mobile and keep all the look and feel (font, formulas, pictures, tables) yet formatted for a smaller screen.
It’s using a local YOLO detector trained specifically on detection pdf page regions.
The old version works and has many users who love it to read scientific papers, but its heuristics based and was in my opinion failing on edge cases that this new AI approach solves.
I created QuickScre because I wanted a no editing way of recording polished screen recordings for Slack etc. Free to try https://www.appblit.com/quickscreen
I grew tired of endless YouTube videos, X articles or web articles.
So this app lets you open any link and you instantly get an AI summary + brief about the content.
(It's free up to 20 articles because there are real costs: I use Gemini to summarize the pages you open)
AI voices run locally on your iPhone/iPad (web extension version coming soon).
It doesn't have to be. You can configure your bot to great the user.
E.g. "Aleksandra is not available at the moment, but I'm her AI assistant to help you book a table. How may I help you?"
So you're telling the caller that it is an AI, and yet you can have a pleasant background audio experience.
I am not using speech to speech APIs like OpenAI, but it would be easy to swap the STT + LLM + TTS to using Realtime (or Gemini Live API for that matter).
OpenAI realtime voices are really bad though, so you can also configure your session to accept AUDIO and output TEXT, and then use any TTS provider (like ElevenLabs or InWord.ai, my favorite for cost) so generate the audio.
Yes DO let you handle long lived websocket connections.
I think this is unique to Cloudflare. AWS or Google Cloud don't seem to offer these things (statefulness basically).
Same with TTS: some like Deepgram and ElevenLabs let you stream the LLM text (or chunks per sentence) over their websocket API, making your Voice AI bot really really low latency.
I built a voice AI stack and background noise can be really helpful to a restaurant AI for example. Italian background music or cafe background is part of the brand. It’s not meant to make the caller believe this is not a bot but only to make the AI call on brand.
I developed a stack on Cloudflare workers where latency is super low and it is cheap to run at scale thanks to Cloudflare pricing.
Runs at around 50 cents per hour using AssemblyAI or Deepgram as the STT, Gemini Flash as LLM and InWorld.ai as the TTS (for me it’s on par with ElevenLabs and super fast)
In browser transcript beautification using a mix of small models (Bert, all-MiniLM-L6-v2 and T5) for restoring punctuation, finding chapter splits and generating the headers.