Alejandro Gómez — NotesNotes on Invisor, agents and software development.https://alejandro.gomezcerezo.com/enReaching an audience that already existshttps://alejandro.gomezcerezo.com/notes/reach-an-existing-audience/https://alejandro.gomezcerezo.com/notes/reach-an-existing-audience/Why I am testing collaborations with real estate creators to bring users to Invisor, with Hermes helping me research profiles and find contacts.Mon, 05 Oct 2026 00:00:00 GMTLearning a price and explaining a pricehttps://alejandro.gomezcerezo.com/notes/discard-before-you-analyze/https://alejandro.gomezcerezo.com/notes/discard-before-you-analyze/At Invisor I have tested CatBoost and Gemini for property valuation. Those comparisons led me to separate data extraction, numerical prediction and explanation.Wed, 26 Aug 2026 00:00:00 GMTThe context that turns a chat into an assistanthttps://alejandro.gomezcerezo.com/notes/autonomy-should-follow-reversibility/https://alejandro.gomezcerezo.com/notes/autonomy-should-follow-reversibility/Hermes saves its activity in Obsidian through a daily cron and links between notes. That memory lets me return to earlier topics and work with accumulated meeting, task and product context.Thu, 20 Aug 2026 00:00:00 GMTThe data split changes what training proveshttps://alejandro.gomezcerezo.com/notes/cost-and-observability-are-product-features/https://alejandro.gomezcerezo.com/notes/cost-and-observability-are-product-features/A model can work on new properties in a familiar area and fail in a different market. My CatBoost experiments test both situations.Wed, 12 Aug 2026 00:00:00 GMTHow much freedom I give the LLMhttps://alejandro.gomezcerezo.com/notes/constrain-behavior-before-generating-language/https://alejandro.gomezcerezo.com/notes/constrain-behavior-before-generating-language/At Zaly I constrained next-step selection with deterministic playbooks. The model retained freedom to write within instructions adapted to the lead.Wed, 05 Aug 2026 00:00:00 GMT