AI Fragrance Formulator
Welcome.
Prototype by SAMARTH GURUDEO HATTALI · Uses the Excel chemical sheet for names, CAS numbers and INR/kg costs.
Application evidence guides Python Formula Architect V2.2. Unknown architecture remains uncertain. Fine Fragrance EDP explicitly selects Fine Fragrance.
Loading local application profiles…
Budget is entered per batch below: INR 60 for 100 g equals INR 600/kg. Performance criteria are formulation goals; application performance has not been measured.
Lower values favour cost efficiency. Higher values allow more premium materials, more naturals, and stronger base support while still respecting dosage limits.
Mode 1 is the normal Formula tab. Search modes generate a reconstruction below. Mode 5 restricts ingredients to those in the availability sheet.
History is stored in browser localStorage.
Generates an EDP-style formula with a deep base pyramid (Top 20% / Middle 30% / Base 50%) and a minimum luxury level of 80/100. Uses the same batch size from the Formula tab. Auto-balances every requested accord above 2%.
Ingredient restriction follows the Yes/No choice in the Formula tab.
Not yet loaded.
Legacy browser model controls. Normal generation uses Python V2.2 with optional configured Ollama review. Model downloads require an explicit action.
Checking browser AI support…
Review rounds
Generate starts with a local Formula Architect draft using your brief, application goals, and available materials. Specialist engines use the shared local knowledge base. Formula Engine checks the draft; DeepSeek independently reviews or edits it. Modify Formula adjusts the current trial. Final Engine validation is required. Workflow diagnostics show stage timings, inferences, model loads, swaps, and rounds. Maximum three rounds; no automatic backend fallback or internet research.
Company Ingredient Knowledge & Trial Comments
Research pending: loading…
Opening local IndexedDB knowledge…
Identity, evidence and completeness
Research runs only when requested. The optional backend fetches public sources; knowledge stays in this browser. Export backups for recovery on another device.
Internal observations remain separate from external evidence. Unverified external identity claims need review before formulation uses them.
Opening local Formula Library…
Opening local Company Learning…
Opening local ingredient intelligence…