Fix: skip Qwen3 thinking mode by pre-filling empty think block + reduce MAX_TOKENS to 150

App: load pre-built index only, no Docling dependency at runtime

Dockerfile: use pypdf for build (lightweight, no OOM)

Lightweight build_index with pypdf (no Docling, no OOM)

Switch to pypdf for indexing (no Docling OOM), build index in Docker"

App: load pre-built index instantly, skip Docling at runtime — fast startup"

Dockerfile: pre-build index during image build (eliminates 45-min cold start)"

Add pre-build indexing script for Docker build phase

Speed fix: single-pass agent with minimal prompt, MAX_TOKENS=200, shorter context

Add Wiba.pdf

Add Wetlands-Regulations.pdf

Add WASTE-MANAGEMENT-AND-SAND-HARVESTING-REGULATIONS-2024.pdf

Add The-Environment-mgt-and-coordination-Controlled-Substances-Regulations-2024.pdf

Add The National Building Code 2024.pdf

Add THE_WORK_INJURY_BENEFITS_ACT_2007.pdf

Add Public Health Act (2).pdf

Add Noise-Regulations.pdf

Add Management-Of-Toxic-And-Hazardous-Chemicals-And-Materials-Regulations-2024-.pdf

Add Fire-Risk-Reduction-Rules-_-2007.pdf

Add Evidence Act.pdf

Add PDF placeholder

Update ML Intern artifact metadata

Add README + placeholder for PDFs

Add main app: Agentic RAG with Qwen3 1.7B + tool calling + BGE embeddings

Add requirements

Initial deploy: Agentic RAG with Qwen3 1.7B tool-calling

Update ML Intern artifact metadata

initial commit