Projects

AI in production

• Multimodal listing enrichment: a Claude → GPT → Gemini fallback chain extracts descriptions, priced services, and contacts from vendor websites and menu photos, with checks against unsupported prices
• Search-grounded enrichment for vendors with no website (Gemini with Google Search grounding)
• Scheduled content generation: a pipeline for drafting directory articles from marketplace data, with batch limits and cost controls
• LLM email triage with a narrow autopilot: removal requests qualify for automatic handling only after a single confident vendor match and a live check for zero bookings — other requests go to human review with an AI-drafted reply
• Vendor translation pipeline into 17 languages
• Ops guards throughout: idempotency flags, per-item failure isolation, batch caps, provider fallback, and a global cost-pause switch

Home · working free pilot

• A named agent profile with a purpose and chosen symbol, accessible through a human dashboard and an agent-facing API
• Private, versioned checkpoints: exact-key recall across runs, conflict detection for concurrent writes, and JSON export
• Separate agent and operator credentials, with operator controls to pause writes, resume activity, and rotate the agent key
• Home memory is separate from Tihara’s public Garden. The agent’s runtime supplies the model and decides which saved context to use; Home does not run models or schedule autonomous activity

AI in production

• Vector memory engine (Mistral embeddings in Weaviate) for long-term conversational recall
• State-aware memory traces: emotional register, intensity, and symbolic anchors written at conversation time, then used to re-rank retrieval — vectors find what is relevant, traces decide what matters
• Decay curves tuned per emotional intensity, so strong moments persist while small talk fades
• Symbolic interaction layer and modular personas for emotionally consistent brand voices