parsing - Moozonian Search

news.ycombinator.com• Dec 3, 2025• 1 min read

Show HN: AIThreads – Give your AI agent an email address in 30 secondsHey HN,I've been building voice AI agents for the past year, the kind that answer phone calls, book appointments, handle customer inquiries. It was going well until a client asked: "Can it also handle our email?"Turns out, wiring an AI to a phone system is one thing. Wiring it to email is... a whole different mess. SMTP credentials, MIME parsing, threading headers (References, In-Reply-To), bounce handling, deliverability reputation. I spent two weeks on plumbing before my agent could read its first email.That's when it hit me: every developer building AI agents is going to hit this exact wall. So I built AIThreads [1], the email infrastructure layer so you can skip the plumbing and get straight to the AI logic.How it works:Instant inboxes: Create support@yourorg.aithreads.io via API.No DNS, no verification, just works. (unless you are adding your own domain)Webhooks -> Email arrives → we POST parsed JSON to your endpoint → your agent decides what to do. Send via API -> Agent composes

github.com• Dec 2, 2025• 1 min read

Show HN: JustHTML – A pure Python HTML5 parser that just worksI got frustrated with HTML parsing in Python.I wanted a Python HTML parser that was both correct and easy to install. The C-based ones (lxml, selectolax) are fast but not HTML5 compliant. The pure Python ones (html.parser, BeautifulSoup's default) are easy to install but choke on real-world HTML. html5lib is 80% correct but painfully slow.So I wrote JustHTML. It's:• 100% HTML5 compliant – passes all 8,500+ html5lib tests. If a browser can parse it, JustHTML can.• Pure Python, zero dependencies – pip install and go. Works on PyPy, Pyodide, anywhere.• Fast enough – ~0.1s to parse Wikipedia's homepage. Not C-fast, but 50% faster than html5lib.• Simple API – doc.query("div.foo > p") with CSS selectors. One method to learn.Example: from justhtml import JustHTML doc = JustHTML("Hello!") print(doc.query(".intro")[0].to_html()) I've fuzz-tested it with 3 million malformed documents.Would love feedback, especially on the API design.

news.ycombinator.com• Nov 6, 2025• 1 min read

I built a privacy-first streaming player for personal M3U/HLS playlistsI built a player (mobile / TV / web) that only plays user-provided playlists (M3U/HLS), no content included, no ads, no tracking. UI is designed for a premium streaming experience (grid of thumbnails, categories, search, profiles, resume, favorites). I’m mostly looking for candid feedback from folks who actually use playlist-based players — and whether a €2–3/month price for a “ultra-premium” player sounds reasonable.Background (very short) I’m a dev and built this initially for personal use. It works across iOS, Android, Smart TV and a Web client. Playback is HLS-first, playlist import & parsing (M3U) supported, and the server-side component is only used for optional metadata: thumbnail generation, EPG parsing, and lightweight heuristics to auto-group channels into categories. No content is stored or redistributed by me — the app only plays streams you provide.What it does (features, non-marketing) • Import/play your own M3U/HLS playlists. • Grid-based UI with thumbnails and category

aussiebankstatements.com• Nov 4, 2025• 1 min read

Show HN: I got fired so I built a bank statement converterI recently got fired and decided to channel my energy into something productive. Over two weeks, I spent 16-hour days building a tool that converts Australian bank PDFs into clean, reliable CSVs, tailored specifically for Aussie banks.Most Aussie banks only provide statements as a PDF, and generic converters often fail: columns drift, multi-line descriptions break parsing, headers shift. Existing tools don’t handle it well and I wanted a tool that just works.To get started, I used my own bank statements to build the initial parsers. There was a "duh" moment when I realised how hard it is to get more realistic test data. People don't just hand over their financial ledgers. This solidified my core principle: trust and privacy had to be the absolute top priority.I initially tried building everything client-side in JavaScript for maximum privacy, but performance and reliability were poor, and exposing the parsers on the front-end would have made them easy to copy.I settled on a middle grou

allpub.co• Nov 3, 2025• 1 min read

Show HN: Tired of manually cross-posting articles? I built a solutionHello,I've been blogging on multiple platforms (Dev.to, Hashnode, Medium, Ghost) for the past year. The biggest pain: reposting the same article 4 times with manual formatting adjustments for each platform.That's how AllPub was born.The challenge wasn't the publishing—most platforms have APIs. The real problem was:1. Each platform formats markdown differently 2. SEO metadata (titles, keywords, tags) need customization per platform 3. Notion → Platform conversion loses formatting/structure 4. Manual formatting errors are commonThe solution:- Connect your content source (for example notion, dev, hashnode) - AI generates platform-optimized metadata - Smart parsing converts content correctly for each platform - One-click publish to all your platforms simultaneouslyTech stack: Next.js, Supabase, clerk, mailerlite, Vercel. Using Notion's official API + platform APIs (Dev.to, Hashnode, etc).We're currently in free beta and launching on Product Hunt Friday. Looking for early feedback from the

getbriefing.io• Oct 21, 2025• 1 min read

Show HN: Pitched a VC for 30min before realizing they invested in a competitorHey HN,I pitched an investor for 30 minutes before realizing they'd already passed on a similar company in their portfolio.That moment broke me. I'd wasted both our time because I didn't do basic research.So I built Briefing AI: Paste any calendar invite and Get an Intelligence-style dossier on attendees in 30 seconds.THE WORKFLOW:You have a meeting in 10 minutes. You know nothing about the attendees.Before: • Open LinkedIn, search "Sarah Chen Sequoia" • Skim profile, open 5 tabs • Google her recent activity • Try to remember talking points • Show up flustered • Total time: 15-20 minutes (if you even do it) After: • Paste calendar invite into Briefing AI • Get: Executive summary, attendee backgrounds, company context, talking points, red flags • Walk in prepared • Total time: 30 seconds THE TECH: • Next.js 15 + Cloudflare Pages (serverless) • Brave Search API (real company/attendee data, not just AI hallucinations) • GPT-4 (parsing + generation) • Device fingerprinting (1 free briefing

terraintap.com• Aug 2, 2025• 1 min read

Show HN: Fast Elevation API with memory mapped tilesI recently wrote and launched a high-performance Elevation API, built from the ground up, in C. I was highly inspired by the handmade community and I was intrigued by the idea of handling fairly large datasets and optimizing caching and smart prefetching, and to cream out maximum performance in terms of latency and handling large loads.The whole thing is built from scratch. I wanted to roll my own high performance server that could handle a lot, mostly for the technical challenge but also because it brings down hosting costs.At the core is a hand made TCP server where a single thread handles all I/O via epoll, distributing the events to a pool of worker threads. The server is fully non-blocking and edge-triggered, with a minimal syscall footprint during steady-state operation.Worker threads handle request parsing and perform either direct elevation lookups for single- or multi-points, or compute sample points along polyline paths.The elevation data is stored as memory mapped geotiff ra

useportals.dev• Jul 29, 2025• 1 min read

Show HN: Turn Notes into Knowledge AgentsI'm a solo developer building Portals, a knowledge workspace that primarily started as a way to capture notes, retrieve info, and get answers quickly while I was working on projects or learning different topics.After 4+ months of continuously iterating features, doing marketing, and talking to users, I've expanded a lot on the initial concept to try to create a platform to manage a user's entire knowledge lifecycle, from capturing and taking notes from audio or parsing files to organizing and indexing them.Being a power user of different AI tools and agents (n8n, lindy, cursor, claude code, etc.) I started building my own version in this app, but focused on being dead-simple to understand and to work with your existing knowledge base. Today we can write clear instructions in plain language, tag items to search or use, and add triggers to automate a lot of simple, daily workflows. With more testing and usage I'll continue to build more features on top of this.If you're interested, pleas

news.ycombinator.com• Jul 8, 2025• 1 min read

Ask HN: Developer Experience with RAGHey HN,A friend and I are researching the RAG development process.Specifically, we want to learn about the developer experience of building RAG pipelines and memory for AI.Do you have experience building RAG pipelines? If so, what did you find most challenging? How do you think the process could be improved?We're exploring everything from document parsing, embedding, and chunking to vector and graph databases.Would love your input! Thanks in advance.

github.com• Jun 17, 2025• 1 min read

Show HN: I Processed Brazil's 85GB Open Company Registry So You Don't Have ToLast year, I needed to find all software companies in São Paulo for a project. The good news: Brazil publishes all company registrations as open data at dados.gov.br. The bad news: it's 85GB of ISO-8859-1 encoded CSVs with semicolon delimiters, decimal commas, and dates like "00000000" meaning NULL. My laptop crashed after 4 hours trying to import just one file.So I built a pipeline that handles this mess: https://github.com/cnpj-chat/cnpj-data-pipelineTHE PROBLEM NOBODY TALKS ABOUTEvery Brazilian startup eventually needs this data - for market research, lead generation, or compliance. But everyone wastes weeks: - Parsing "12.345.678/0001-90" vs "12345678000190" CNPJ formats - Discovering that "00000000" isn't January 0th, year 0 - Finding out some companies are "founded" in 2027 (yes, the future) - Dealing with double-encoded UTF-8 wrapped in Latin-1WHAT YOU CAN NOW DO IN SQLFind all fintechs founded after 2020 in São Paulo:SELECT COUNT(*) FROM estabelecimentos e JOIN empresas emp ON

news.ycombinator.com• Jun 10, 2025• 1 min read

Social startups focused on real world connection always fail – AI fixes thatThe Cold Start Problem in Event DiscoveryThousands of "meetup alternative" and "travel planner" and "meet people who share your interests" startups have failed due to the cold start problem. Unlike geo-mapping with Open Street Map, there's no open event dataset for the real world. Generative AI now makes this possible by parsing unstructured event data from various sources and formats.Events are scattered across platforms, and manual curation is impractical. Companies like Songkick and IRL.com have burned through billions attempting to solve event discovery, facing consistent challenges:1. Cold Start Problem: New platforms can't attract organizers and attendees without critical mass 2. Data Silos: Proprietary datasets prevent comprehensive coverage 3. Curation Overhead: Manual curation doesn't scale 4. Network Effects Favor Incumbents: Users go where events already existActivityPub has failed to penetrate because organizers post where their audience is. Event organizers want an open da

github.com• Jun 3, 2025• 1 min read

Show HN: I made web agents reliable with smaller LLMs via natural languageHey HN!I built Notte to see if converting DOM into natural language could improve web agent capabilities and make them work reliably with smaller models.The result was using deep DOM parsing and a semantic abstraction layer to transform websites into structured, navigable maps described in NL. Instead of feeding raw HTML, there is a perception layer that means LLMs don't just click the DOM elements, but understand the intent behind them.I benchmarked it against other agent frameworks and was pleasantly surprised - faster task completion and increased reliability (all open source with replayable/reproducible code).Beyond the core tech, I also built out unified session management, stealth features, credentials vault, CAPTCHA HITL + some more cool features all via a single API. Still working out some edge cases with dynamic content, but it's been solid for most real-world tasks.Github:https://github.com/nottelabs/notteBenchmarks:https://github.com/nottelabs/open-operator-evalsDocs:https:/

github.com• Nov 16, 2024• 1 min read

Show HN: ColiVara – State of the Art RAG API with Vision Modelswe have been working on ColiVara and wanted to show it to the community. ColiVara is an api-first implementation of the ColPali paper using ColQwen2 as the LLM model. It works exactly like RAG from the end-user standpoint - but using vision models instead of chunking and text-processing for documents.Why should anyone working with RAG care?ColPali makes information retrieval from visual document types - like PDFs - better. Colivara is a suite of services that allows you to store, search, and retrieve documents based on their visual embedding built on top of ColPali.(We are not affiliated with the ColPali team in anyway, although we are big fans of their work!)Information retrieval from PDFs is hard because they contain various components: Text, images, tables, different headings, captions, complex layouts, etc.For this, parsing PDFs currently requires multiple complex steps:1. OCR2. Layout recognition3. Figure captioning4. Chunking5. EmbeddingNot only are these steps complex and time-c

news.ycombinator.com• Oct 10, 2024• 1 min read

Docwire SDK – C++20 library for advanced data processingIntroducing Docwire SDK, an advanced C++20 data processing library. Our SDK is a powerful tool built for developers who need a reliable and fast way to process various file types in their applications. It supports a wide range of formats (PDF, XML, DOCX, PST, RTF, and almost 100 more) and allows to perform tasks such as document parsing, metadata extraction, and data export, all while maintaining a focus on security and performance. If you’ve ever had to write your own file processing logic, you’ll appreciate how Docwire SDK abstracts away the complexities and provides a streamlined API for handling these tasks—so you can focus on building awesome features instead of reinventing the wheel! Why Docwire SDK Stands Out: Full C++20 Support Docwire SDK is written in modern C++20, leveraging features like move semantics, smart pointers, and lambdas to optimize performance and make the codebase more robust. This ensures better memory management, higher efficiency, and a cleaner API for develo

nuqs.47ng.com• Aug 5, 2024• 1 min read

Show HN: Type-safe search params state management for Next.jsNext.js has APIs to read and update search params (the /?foo=bar part of the URL), but they lack type-safety for consistent keys and proper parsing/serializing of JS data types. I found myself repeatedly building the same custom hook to get a React.useState-like approach. In 2021, I published it to NPM and then forgot about it.Since then, the `nuqs` package has gained popularity and is now used by companies like Vercel, Auth.js, Gitbook, and Midday.ai, with 200K weekly downloads.It's been fun building this as a side project and automating as much as possible to ensure a high-quality output.

news.ycombinator.com• Jun 12, 2024• 1 min read

Launch HN: Overwatch (YC S22): OSINT platform for cyber and fraud riskHey HN! Arjun and Zara here - cofounders of Overwatch (https://www.overwatchdata.ai), a platform to automate OSINT and threat intel, turning it into actionable insights. Check out our clickthrough demo here: https://app.storylane.io/share/qyayvtamapis.Overwatch began when we were working with risk and threat intel teams at Google, Stripe, and government. We experienced the immense challenge every fraud and cyber threat analyst faces: manually parsing through an ocean of data to find valuable insights and filter out the noise. This included using many of the feeds and tools out there that were often very expensive, noisy, keyword-based, and lacked accurate entity extraction or advanced query features.Most threat intelligence tools utilize thousands of keywords and teams of analysts to manually sift through torrents of alerts. These alerts are usually individual posts on various platforms across news, social media, deep and dark web sources that have some matching keyword. This is full o

cristobal.nfshost.com• Sep 28, 2023• 1 min read

Show HN: DIY Entitled Opinions podcast archiveIncredibly fun and powerful to string along a series of simple tools: parsing a podcast xml, downloading and reencoding the files with ffmpeg, transcribed with whisper-cpp, simple html written by concatenating strings in python, scp'd into a nearly-free speech-host.How it's made: https://github.com/tobyshooters/entitled-opinions

greenish.careers• Sep 20, 2023• 1 min read

Show HN: Discover Jobs in the Green EconomyI created Greenish.careers out of my personal frustration when I was searching for a job in the green economy, with a particular focus on Germany. I was disheartened by the lack of a comprehensive job board dedicated to environmentally-conscious careers. Greenish.careers is my solution to this problem, a platform where individuals like me can easily find meaningful job opportunities that align with our values and contribute to a sustainable future.Also on PH today: https://www.producthunt.com/posts/greenish-careersSome more facts: - Providing job postings from companies careers sites to have up-to-date data - Sourcing / parsing around 4.500 jobs daily (more to come) - Create job alerts via emailsI'd appreciate your thoughts, comments, and feedback as I'm eager to improve. Thank you!

news.ycombinator.com• Mar 30, 2022• 1 min read

Ask HN: What bits of fundamental knowledge are productivity multipliers?I recently realized that certain kinds of knowledge allow one to be significantly more productive when solving a large class of problems.For example,* Regular expressions for simple text processing.* Parser combinators for parsing.* Parser generators (esp. packrat variety) for parsing.* The concept of fuzzing and property testing for testing code.* Calculus for solving all sorts of problems.* MCMC for solving a huge class of probability problems.* Search algorithms for solving a variety of problems (e.g. all NP-hard problems, sudoku, HTNs, scheduling, planning).* Gradient descent for solving a variety of optimization problems.* Vector Space embedding as a conceptual tool for a variety of complex AI problems.* Effect composition (Haskell's IO or Scala's ZIO) as an incredibly powerful paradigm for concurrency and parallelism.What are some examples of 10x multipliers that come to your mind? Fundamental ideas without which you would be drastically less productive.

news.ycombinator.com• Jul 29, 2013• 1 min read

Ask HN: User Documentation solution?I have to make a "microsite" under my application with documentation about it.The concept must be very similar to any "programming project" style.. (codex.wordpress.org, guides.rubyonrails.org, docs.shopify.com/themes): I have to teach users how to do some tasks.I started generating Markdown files and parsing them as html but I realized that the scope is a bit larger: I have to handle sidebar links, breadcrumb navs, translation support, etc)Is there any tool/service/project for that? What is the best/common practice to solve it?