A local-first inventory management system for Nepali retail shops. Shop owners can manage stock using Nepali voice commands, view sales insights, and ask an AI assistant about their business.

Small retail shops often manage inventory manually, making it difficult to keep stock records accurate, identify low-stock products, and understand sales performance.Manual inventory makes stock tracking, low-stock detection and sales analysis difficult.
Bikri AI combines Nepali voice commands, real-time inventory management, sales reporting and an AI business assistant into one local-first system.One local-first system combining Nepali voice inventory, sales reports and an AI business assistant.
Building the system taught me how speech recognition, LLMs, semantic product matching, vector search and traditional backend logic can work together in a real application.I learned how speech recognition, LLMs, vector search and backend logic can work together in a real application.





A Nepali voice command is transcribed locally with Whisper, interpreted into an inventory action, matched against products using semantic search, and processed through the FastAPI backend and PostgreSQL database.
Voice input, inventory and chatbot UI
Speech processing, AI logic and API
Products, transactions and vector data
Voice input, inventory and chatbot UI
Speech processing, AI logic and API
Products, transactions and vector data
Bikri AI combines Whisper, Llama 3, Qwen 2.5 and SBERT to provide voice inventory control, semantic product matching and AI-powered business analysis without cloud APIs.Local AI powers voice inventory, product matching and business analysis without cloud APIs.

Next project