KACM

Dial is a privacy-first semantic zoom reader for sensitive documents. It lets users move through the same document at different levels of complexity, from the original expert text to standard, plain, and simple language, using an interactive dial. The goal is to help people understand administrative letters, contracts, and technical documents without uploading private information to the cloud. Dial runs locally with Gemma through Ollama. At ingest, it generates all reading levels once, caches them, and then makes switching levels instant. To preserve trust, Dial detects hard facts such as amounts, dates, deadlines, percentages, names, and legal references, seals them as locked facts, and requires them to remain unchanged across every rewrite. A second-model audit layer using Nemotron checks whether the simplified text may have changed the meaning, and flags uncertain passages instead of pretending the rewrite is always safe. The result is not a chatbot and not a basic document summarizer. Dial is a verifiable reading instrument: it gives users control over document complexity while keeping sensitive text local, preserving critical facts, and showing when the AI might have betrayed the original meaning.