Hallucina-Gen - Spot where your LLM might make mistakes on documents

Hi fellow makers, We’re Actualization.AI, and our mission is to make AI safer, more accurate, and genuinely helpful to people. We built this tool for AI developers or tinkerers who are using LLMs to work with documents, especially those building things like policy chatbots, legal assistants, or internal tools over PDFs. LLMs are powerful, but they’re also prone to hallucinating. This is especially true when summarizing or answering questions from dense documents like regulations, contracts, or technical reports. We see this all the time in our own AI projects, and we wanted a way to catch those errors before they reach users. This project started with our own frustrations. One of us is a professor and AI researcher at USF. The other has a PhD specializing in AI/NLP. We kept running into the same issue: LLMs giving confident, wrong answers in high-stakes domains. So we built a tool to make AI development more robust. Here’s how it works: - Upload your PDFs - Tell us what LLM you're using and the prompts you're testing - We analyze the documents and generate a spreadsheet of inputs likely to trigger mistakes It helps you test prompts, improve guardrails, and understand how your model might fail. You don't need to run live model calls or set up complex eval pipelines. This is primarily for: - Developers building AI tools that reason over documents - Teams deploying RAG systems or internal copilots - Researchers and QA folks focused on LLM evaluation and reliability - People just tinkering with chatbots that want to know how accurate they are We’re providing Product Hunt users a free preview of the hallucinations we find, and we only ask for one thing: We’d love feedback from early users. What’s helpful? What’s missing? How can we make this better? Thanks for checking it out. We’re here all day if you have questions or want to go deeper. - The Actualization.AI team