TimeLens: On-Device Artifact Recognition with Retrieval-Augmented Question Answering for the Grand Egyptian Museum

2026-06-11 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

TimeLens is an AI-powered bilingual mobile guide developed for the Grand Egyptian Museum (GEM), enabling visitors to point their phone at an exhibit for real-time artifact recognition and follow-up questions in English or Arabic. This system addresses challenges like fine-grained visual similarity among 51 catalogued artifacts, the disparity between curated training data and handheld camera conditions, and the need to prevent unsupported historical claims. Its engineering contributions include an on-device artifact detector, a 5.97 MB TensorFlow Lite YOLOv8n model, achieving mAP@0.5 = 0.995 and mAP@0.5:0.95 = 0.924, developed through a data-quality-driven iteration process. Additionally, a bilingual Retrieval-Augmented Generation (RAG) guide, grounded in a 108-record ChromaDB knowledge base, utilizes Gemma 4 E2B (Q4 K M) and incorporates ten optimizations to reduce end-to-end latency from over 30 seconds to approximately 10 seconds. Both subsystems are integrated into a production Flutter application featuring a bilingual interface, museum location gating, and text-to-speech support.

Key takeaway

For AI Engineers developing on-device AI applications for complex visual environments or real-time Q&A, prioritize data quality and iterative refinement in your vision model development. The TimeLens project demonstrates that meticulous label quality is decisive for achieving high accuracy with efficient models like YOLOv8n. Additionally, implement targeted optimizations for Retrieval-Augmented Generation (RAG) systems to reduce end-to-end latency, ensuring a responsive user experience. Consider a hybrid architecture combining on-device vision with a grounded RAG knowledge base for factual accuracy and bilingual support.

Key insights

On-device artifact recognition and RAG can deliver real-time, accurate, bilingual museum guides, overcoming data and latency challenges.

Principles

• Label quality is decisive for on-device object detection.
• Iterative data-quality-driven development improves model performance.
• Targeted optimizations significantly reduce RAG system latency.

Method

Develop on-device detectors via data-quality iteration (auto-annotation, label-cleaning, hand-annotation). Integrate with a RAG system using a local knowledge base and optimized LLM for bilingual Q&A.

In practice

• Use YOLOv8n for efficient on-device object detection.
• Ground RAG with a ChromaDB knowledge base for factual accuracy.
• Optimize LLM inference for mobile RAG latency reduction.

Topics

• On-Device AI
• Artifact Recognition
• Retrieval-Augmented Generation
• Computer Vision
• Mobile Applications
• Grand Egyptian Museum

Best for: Computer Vision Engineer, NLP Engineer, AI Scientist, AI Engineer, Machine Learning Engineer, AI Architect

Related on AIssential

• Technical advances in document understanding — Document processing has evolved from basic OCR to sophisticated multimodal models that preserve structure and fine detail.
• Vector Search Using Ollama for Retrieval-Augmented Generation (RAG) — RAG combines vector search with LLMs to provide context-aware, fact-grounded responses from external data.
• Introducing V-RAG: revolutionizing AI-powered video production with Retrieval Augmented Generation — V-RAG enhances AI video generation by integrating image retrieval, offering customization and accuracy without model retraining.
• Fast and Faithful: Real-Time Verification for Long-Document Retrieval-Augmented Generation Systems — Full-context verification for RAG systems significantly improves hallucination detection in long documents under real-time constraints.
• From Scenes to Elements: Multi-Granularity Evidence Retrieval for Verifiable Multimodal RAG — Element-level visual evidence retrieval significantly enhances multimodal RAG verifiability and performance.
• LAST-RAG: Literature-Anchored Stochastic Trajectory Retrieval-Augmented Generation for Knowledge-Conditioned Degradation Model Selection — LAST-RAG integrates observed data with domain knowledge for robust degradation model selection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.