Docling RAG

[tbrandenburg]

@task Turn this into a simple UV project: https://colab.research.google.com/github/docling-project/docling/blob/main/docs/examples/rag_langchain.ipynb
and test it with a dummy PDF.
Keep it very simple.
Run it without an HuggingFace token.
I want to run it with "uv run docling-milvus-rag"

[github-actions]

🌿 Branch created: task-machine-36

[github-actions]

Context

The team must turn the Docling RAG Colab example into a minimal UV-managed Python project that can be executed locally with uv run docling-milvus-rag, avoids any HuggingFace token requirements, and proves the workflow with a dummy PDF input end to end.

Role Descriptions

business-analyst

• Agent Path: .github/agents/08-business-product/business-analyst.md

llm-architect

• Agent Path: .github/agents/05-data-ai/llm-architect.md

backend-developer

• Agent Path: .github/agents/01-core-development/backend-developer.md

build-engineer

• Agent Path: .github/agents/06-developer-experience/build-engineer.md

cli-developer

• Agent Path: .github/agents/06-developer-experience/cli-developer.md

test-automator

• Agent Path: .github/agents/04-quality-security/test-automator.md

Chronologic Task List

Feature 1 – Docling RAG pipeline adaptation

• [business-analyst] Requirement synthesis — Review output/36/issue_conversation.md alongside the referenced Colab notebook to capture explicit functional/non-functional needs (Docling ingestion, Milvus usage, dummy PDF test, no HuggingFace token) and save a prioritized list with acceptance criteria to output/36/rag_requirements.md.
  • Summary: Consolidated Docling→Milvus requirements into prioritized FR/NFR list with offline/UV constraints and acceptance criteria in output/36/rag_requirements.md.
• [llm-architect] System architecture — Read output/36/rag_requirements.md, inspect the Colab notebook code path, and design a modular Docling→chunking→embedding→Milvus→LLM answering flow covering component contracts, dependency choices, and configuration defaults; record diagrams and dataflow notes in output/36/rag_architecture.md.
  • Summary: Captured Docling→Milvus offline architecture with component contracts, dependency/LLM choices, config defaults, and flow diagrams in output/36/rag_architecture.md.
• [backend-developer] Module design — Using output/36/rag_architecture.md, define Python package layout (e.g., src/docling_milvus_rag/pipeline.py, config.py, milvus_store.py), configuration loading strategy, and interfaces for ingesting local PDFs without remote credentials; document this detailed plan in output/36/rag_module_design.md.
  • Summary: Captured package layout, config layering, and offline ingestion interfaces in output/36/rag_module_design.md.
• [backend-developer] Pipeline implementation — Follow output/36/rag_module_design.md to build the Docling RAG modules under src/docling_milvus_rag/, wire Milvus client initialization that works offline, ensure embeddings/models rely on locally accessible providers, add a sample dummy PDF under fixtures/dummy.pdf, and update pyproject.toml plus uv.lock so dependencies resolve via uv run.
  • Summary: Added the full offline-friendly package under src/docling_milvus_rag/, wired caching + Milvus Lite fallback, created defaults/dummy PDF fixtures, and introduced pyproject.toml with a pinned uv.lock for uv run docling-milvus-rag.
• [test-automator] Pipeline verification — Create tests/test_rag_pipeline.py that loads fixtures/dummy.pdf, runs the pipeline entrypoint, and asserts vector store population plus answer generation; capture execution logs, sample outputs, and pass/fail status in output/36/rag_test_report.md.
  • Summary: Added pytest-based pipeline coverage, captured sample run evidence, and logged the passing results in output/36/rag_test_report.md.

Feature 2 – UV CLI packaging & execution experience

• [business-analyst] CLI requirement brief — Analyze output/36/rag_requirements.md plus any stakeholder notes to describe expected CLI ergonomics (flags for Milvus URI, paths for documents, output formatting) and document them with usage scenarios inside output/36/cli_requirements.md.
  • Summary: Captured offline-first CLI flags, configuration layering, and usage scenarios with acceptance criteria in output/36/cli_requirements.md.
• [build-engineer] UV project architecture — Read output/36/cli_requirements.md to define the UV workspace layout, dependency constraints, and reproducible lockfile strategy; specify scripts, pyproject.toml metadata, and uv run command wiring inside output/36/uv_build_architecture.md.
  • Summary: Documented UV workspace layout, dependency/lock strategy, and command wiring in output/36/uv_build_architecture.md so uv run docling-milvus-rag stays deterministic and offline-friendly.
• [cli-developer] Command design — Based on output/36/uv_build_architecture.md, describe the CLI command tree for docling-milvus-rag, list arguments/env vars, outline help text, and determine how results/dummy PDF checks surface; store this plan in output/36/cli_command_design.md.
  • Summary: Captured Typer command tree, flag/env matrix, help text grouping, and dummy PDF validation workflow inside output/36/cli_command_design.md.
• [cli-developer] CLI implementation — Implement the command described in output/36/cli_command_design.md by adding src/docling_milvus_rag/cli.py (and __main__.py if needed), exposing an entry point in pyproject.toml, wiring logging plus error handling, and updating README usage notes so uv run docling-milvus-rag --pdf fixtures/dummy.pdf executes the pipeline.
  • Summary: Delivered Typer-based CLI with default run behavior, preflight/store utilities, README usage docs, and a [project.scripts] entry so uv run docling-milvus-rag --pdf fixtures/dummy.pdf runs the offline pipeline end to end.
• [test-automator] CLI integration tests — Author tests/test_cli_entrypoint.py that shells out via uv run docling-milvus-rag --pdf fixtures/dummy.pdf, captures stdout/stderr, and validates exit codes plus output artifacts; summarize coverage gaps and test evidence in output/36/cli_test_report.md.
  • Summary: Added pytest that shells out to uv run docling-milvus-rag with isolated artifacts, skips gracefully when uv is absent, and documented the current skipped state plus rerun guidance inside output/36/cli_test_report.md.