{
  "id": 12001893,
  "title": "Horizon -Look Back & See Forward.",
  "url": "https://urgent.news/2026/10/04/horizon-look-back-see-forward",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-04T20:46:28.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/harsh_srivastava_bf21360d/horizon-look-back-see-forward-22oi"
  },
  "original_language": "en",
  "account": "This submission, titled Horizon - Look Back & See Forward, addresses the challenge of building for a friend by creating an intelligence layer on top of a personal journal. The journal retains its familiar experience while becoming more useful as the user writes more. Instead of simply asking \"What did I write?\", users can ask \"What have I been repeatedly thinking about?\" or \"What has been taking most of my attention lately?\" The system answers these questions using the history stored in the journal, with each answer linking back to the relevant entries.\n\nThe author built Horizon for themselves, as they wanted their journal to remember ideas, goals, projects, people, interests, and topics that might have been forgotten or repeatedly mentioned. The demo uses a fictional student's five weeks of journal entries to showcase the various features without exposing the author's real journal. Users can ask questions about their journal and receive answers based on relevant past entries. Each answer includes the entries it is based on, along with a label indicating the strength of the evidence. If no entries support an answer, Horizon will indicate that there is not enough evidence.\n\nHorizon identifies recurring ideas, goals, projects, people, and topics across entries, surfacing things that may have been mentioned once and forgotten, as well as ideas that keep coming back. It also tracks unresolved intentions, such as \"I still need to finish my portfolio.\" or \"I should apply for that internship,\" and marks them as resolved, dismissed, or finished when addressed later. All data is kept, preserving the journal's history.\n\nThe author's goal is to make a journal more useful the longer it is used, rather than building another AI chatbot. The demo is already seeded with five weeks of fictional journal entries, ensuring every feature has real history to work with. The live demo can be accessed at https://horizon-theta-vert.vercel.app, with the password being hackmyfriend@7007. Code for the project is available on GitHub at https://github.com/harsh373/Friend.\n\nThe AI system for Horizon processes journal entries in stages, without seeing the whole journal at once. Structured notes and embeddings are extracted from each entry upon saving, and a question's embedding is compared to the stored entry embeddings. Only the closest entries are passed to the open-weight model for generating a grounded answer with links to the original entries. This ensures that Horizon remains honest and avoids making confident guesses in the journal, which is not the appropriate place for such information. The model's prompt includes checks to ensure every extracted item includes a word-for-word quote from the entry, and source entries only come from the database. The model only picks entry numbers, and the server maps them back to real dates.",
  "summary": "This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend What I Built A journal is good at remembering what happened. But after months of writing, it becomes surprisingly difficult to remember what kept happening. I built Horizon , an intelligence layer on top of a personal journal. The journaling experience stays familiar, and the journal becomes more useful the longer…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}