Memoria — Enterprise Institutional Memory AI

Every enterprise loses knowledge constantly. A senior leader leaves and takes 10 years of context with them. A team makes a decision that contradicts something resolved two years ago in a meeting nobody remembers. The minutes say what was decided — but never why, what alternatives were rejected, or what risks were flagged. Memoria solves institutional amnesia. It processes meeting transcripts through a multi-model Gemini pipeline — using Gemini Flash for intelligent query routing and complexity classification, and Gemini Pro for deep reasoning and structured decision extraction. Every decision is embedded using Gemini's text-embedding-004 model and stored in a ChromaDB vector database, enabling semantic search that finds related decisions by meaning, not just keywords. The system has two modes. Live Meeting Mode processes transcripts in real time, extracting what was decided, why, who proposed it, what alternatives were rejected, and what risks were flagged. Historical Mode absorbs past documents — PDFs, notes, old meeting logs — turning years of buried knowledge into searchable memory. When a team member asks "have we tried this before?" — Memoria retrieves the most semantically relevant past decisions, synthesizes them into a cited answer, and warns if the current direction contradicts a previous lesson learned. The model routing architecture is a core technical differentiator: Flash classifies query complexity first, routing simple lookups to stay on Flash while escalating analytical, multi-step, or comparative questions to Pro. This reduces cost and latency while maintaining quality where it matters. Built for mid-sized enterprises — healthcare networks, financial services firms, manufacturers — where institutional knowledge is a competitive asset and its loss is measurable in dollars. Track: Data & Intelligence. Technologies: Gemini Pro, Gemini Flash, text-embedding-004, Google AI Studio, ChromaDB, FastAPI.