TECHNICAL TEST · AI ENGINEER · DEMO 24/7
Multimodal RAG — Laporan Bank Mandiri 2025
📥 A1 · Pipeline Ingestion (Multimodal)
PDF 9 hal
→
168 section
→
8 caption visual
→
44 chunk
→
embedding MiniLM
→
ChromaDB
📚 A2 · Knowledge Base — hasil parsing multimodal
Semua jenis
📄 Teks
📊 Tabel
🖼️ Visual / caption VLM
💬 A3 · Query — retrieval + sitasi halaman
🔍 Cari di dokumen
✨ + Jawaban LLM
🏆 Hasil evaluasi resmi — 6/6 halaman sumber & substansi benar
🖼️ B · Layout Aware Text Extraction (Computer Vision)
📊 Galeri asli vs hasil
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5