Show HN: I Open-Sourced the Future of Search (manifesto-ao2n.vercel.app)

Alessandro Aimar has open-sourced a prototype Rails 8 AI-powered search assistant that rethinks search as a personalized, truth-grounded experience rather than a generic aggregation of links or model “knowledge.” The repo implements a multi-stage pipeline—web search → scraping → vectorization → AI synthesis—that only uses scraped content to generate answers, enforces inline citations, and explicitly forbids hallucination or injection of LLM prior knowledge. The project frames the core problem not as searching but as delivering relevance tuned to a user’s goals and context, aiming for answers that “just get” the asker. For the AI/ML community the work is notable because it operationalizes several practical guardrails often discussed theoretically: grounding LLM outputs in retrieved documents, mandatory citation, input/output sanitization, timeouts/retries, rate-limiting, and fallbacks for scraping failures. Technically, it highlights trade-offs—scraping reliability, freshness, vectorization choices, and how to encode personalization instructions—while offering a concrete, auditable architecture for reducing hallucinations. The code is an unfinished draft and an invitation to contribute: it’s useful as a living example of building citation-first, goal-aware search agents and a starting point for research and engineering on robust, personalized retrieval-augmented systems.