visual-parser 1.0.0__tar.gz

visual_parser-1.0.0/PKG-INFO

@@ -0,0 +1,191 @@
+Metadata-Version: 2.4
+Name: visual-parser
+Version: 1.0.0
+Summary: Standalone Visual-RAG PDF Parser — text extraction + Vision-LLM figure descriptions → JSONL
+License: MIT
+Project-URL: Homepage, https://github.com/SmartLabNuclear/RADIANT_LLM
+Project-URL: Repository, https://github.com/SmartLabNuclear/RADIANT_LLM
+Project-URL: Docker Hub, https://hub.docker.com/r/zev94/radiant-llm
+Keywords: pdf,rag,nougat,vision-llm,ocr,document-parsing,jsonl,knowledge-base
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Text Processing :: Markup
+Classifier: Intended Audience :: Science/Research
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: PyMuPDF==1.24.7
+Requires-Dist: Pillow==11.1.0
+Requires-Dist: torch==2.7.0
+Requires-Dist: transformers==4.45.2
+Requires-Dist: huggingface-hub==0.36.0
+Requires-Dist: langchain-community==0.3.3
+Requires-Dist: langchain==0.3.13
+Requires-Dist: langchain-text-splitters==0.3.4
+Requires-Dist: openai==1.78.1
+Requires-Dist: google-generativeai==0.8.5
+Requires-Dist: python-dotenv==1.1.0
+Requires-Dist: tqdm==4.67.1
+Provides-Extra: ocr
+Requires-Dist: pytesseract==0.3.13; extra == "ocr"
+Provides-Extra: dev
+Requires-Dist: pytest; extra == "dev"
+Requires-Dist: ruff; extra == "dev"
+Requires-Dist: mypy; extra == "dev"
+
+# visual-parser (Standalone Visual-RAG PDF Ingestion)
+
+<!-- ![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg) -->
+![Python 3.12.10](https://img.shields.io/badge/Python-3.12.10-brightgreen.svg)
+<!-- ![PyTorch](https://img.shields.io/badge/PyTorch-2.0%2B-ee4c2c.svg) -->
+
+`visual-parser` is a standalone document-ingestion tool that converts PDFs into a multi-modal JSONL knowledge base (text chunks + figure descriptions + metadata). The intended workflow is:
+
+1) Run `visual-parser` on curated PDFs to generate JSONL KB files.
+2) Run RADIANT-LLM Visual-RAG for QA over the generated KB.
+
+## Outputs (JSONL KB)
+
+By default, the pipeline writes:
+- `01_chunks_kb.jsonl`: chunked text extracted from PDFs (Nougat by default).
+- `02_figures_kb.jsonl`: figure/page visual descriptions (Vision LLM).
+- `03_metadata_kb.jsonl`: document metadata rows (title/author/etc.).
+- `04_processed_pdfs.txt`: a tracker so re-runs only process new PDFs (unless `--rebuild`).
+
+## API keys (`.env`)
+
+Provide at least one provider:
+- `OPENAI_API_KEY` (OpenAI)
+- `GEMINI_API_KEY` (Gemini)
+
+Optional:
+- `HF_TOKEN` (if you use gated Hugging Face models)
+
+## Run with Docker (Docker Hub)
+
+Prebuilt images are on **[zev94/radiant-llm](https://hub.docker.com/r/zev94/radiant-llm)** under the **visual-parser** tags:
+
+| Tag | Description |
+|-----|-------------|
+| `visual-parser-1.0` | Pinned release |
+| `visual-parser-latest` | Latest visual-parser build |
+
+### 1) Install Docker
+- Docker Desktop (Windows/macOS) or Docker Engine (Linux)
+
+### 2) Pull the image
+```bash
+docker pull zev94/radiant-llm:visual-parser-1.0
+```
+
+### 3) Run (input + output on the same mounted folder)
+Windows PowerShell:
+```powershell
+docker run --rm --env-file .env `
+  -v "C:\path\to\pdfs:/data" `
+  zev94/radiant-llm:visual-parser-1.0 `
+  --input-dir /data --output-dir /data
+```
+
+Linux / WSL:
+```bash
+docker run --rm --env-file .env \
+  -v "/path/to/pdfs:/data" \
+  zev94/radiant-llm:visual-parser-1.0 \
+  --input-dir /data --output-dir /data
+```
+
+### 4) Run (separate output directory)
+Windows PowerShell:
+```powershell
+docker run --rm --env-file .env `
+  -v "C:\path\to\pdfs:/data" `
+  -v "C:\path\to\out:/out" `
+  zev94/radiant-llm:visual-parser-1.0 `
+  --input-dir /data --output-dir /out
+```
+
+### Offline install (legacy `.tar`)
+
+```powershell
+docker load -i .\visual-parser_0.1.0.tar
+docker images   # use the tag printed by Docker
+```
+
+### Model overrides (optional)
+
+Default vision model is **GPT-5.5** when using `--vision-provider gpt`. Override on the command line:
+
+```powershell
+docker run --rm --env-file .env -v "C:\path\to\pdfs:/data" `
+  zev94/radiant-llm:visual-parser-1.0 `
+  --input-dir /data --output-dir /data --vision-model gpt-5.4
+```
+
+<!-- ## Run from source (Python)
+
+From `codebase/Visual-Parser/`:
+```powershell
+python visual-parser.py --input-dir "C:\path\to\pdfs"
+``` -->
+
+## Common configuration flags
+
+After pulling the image, run:
+
+```bash
+docker run --rm zev94/radiant-llm:visual-parser-1.0 --help
+```
+
+For copy-paste **Docker** examples (vision presets, text modes, workers, rebuild), see [`docker-usage-examples.md`](docker-usage-examples.md).
+
+Paths:
+- `--input-dir` / `-i` (required)
+- `--output-dir` / `-o` (default: same as input)
+
+Text extraction:
+- `--text-mode nougat|lightweight` (default: `nougat`)
+- `--nougat-model facebook/nougat-small`
+- `--chunk-size 500`
+- `--chunk-overlap 100`
+
+Vision LLM:
+- `--vision-provider gpt|gemini` (default: `gpt`)
+- `--vision-model gpt-5.2` (or `gpt-4o`, `gemini-2.5-flash`, etc.)
+- `--vision-detail low|high|auto`
+- `--reasoning-effort none|low|medium|high|xhigh`
+- `--metadata-pages 2`
+
+Performance / misc:
+- `--max-workers 4`
+- `--rebuild` (reprocess everything; ignore `04_processed_pdfs.txt`)
+- `--log-level DEBUG|INFO|WARNING|ERROR`
+
+---
+
+## Citation
+
+If you use RADIANT-LLM or the accompanying evaluation materials, please cite the preprint:
+
+```bibtex
+@article{ndum2026radiant,
+  title={RADIANT-LLM: an Agentic Retrieval Augmented Generation Framework for Reliable Decision Support in Safety-Critical Nuclear Engineering},
+  author={Ndum, Zavier Ndum and Tao, Jian and Ford, John and Yim, Mansung and Liu, Yang},
+  journal={arXiv preprint arXiv:2604.22755},
+  year={2026}
+}
+```
+
+Preprint: https://arxiv.org/abs/2604.22755
+
+---
+
+## License
+
+This repository is currently proprietary and not licensed for public use, redistribution, or modification. Licensing terms will be updated after institutional review.
+
+

visual_parser-1.0.0/README.md

@@ -0,0 +1,152 @@
+# visual-parser (Standalone Visual-RAG PDF Ingestion)
+
+<!-- ![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg) -->
+![Python 3.12.10](https://img.shields.io/badge/Python-3.12.10-brightgreen.svg)
+<!-- ![PyTorch](https://img.shields.io/badge/PyTorch-2.0%2B-ee4c2c.svg) -->
+
+`visual-parser` is a standalone document-ingestion tool that converts PDFs into a multi-modal JSONL knowledge base (text chunks + figure descriptions + metadata). The intended workflow is:
+
+1) Run `visual-parser` on curated PDFs to generate JSONL KB files.
+2) Run RADIANT-LLM Visual-RAG for QA over the generated KB.
+
+## Outputs (JSONL KB)
+
+By default, the pipeline writes:
+- `01_chunks_kb.jsonl`: chunked text extracted from PDFs (Nougat by default).
+- `02_figures_kb.jsonl`: figure/page visual descriptions (Vision LLM).
+- `03_metadata_kb.jsonl`: document metadata rows (title/author/etc.).
+- `04_processed_pdfs.txt`: a tracker so re-runs only process new PDFs (unless `--rebuild`).
+
+## API keys (`.env`)
+
+Provide at least one provider:
+- `OPENAI_API_KEY` (OpenAI)
+- `GEMINI_API_KEY` (Gemini)
+
+Optional:
+- `HF_TOKEN` (if you use gated Hugging Face models)
+
+## Run with Docker (Docker Hub)
+
+Prebuilt images are on **[zev94/radiant-llm](https://hub.docker.com/r/zev94/radiant-llm)** under the **visual-parser** tags:
+
+| Tag | Description |
+|-----|-------------|
+| `visual-parser-1.0` | Pinned release |
+| `visual-parser-latest` | Latest visual-parser build |
+
+### 1) Install Docker
+- Docker Desktop (Windows/macOS) or Docker Engine (Linux)
+
+### 2) Pull the image
+```bash
+docker pull zev94/radiant-llm:visual-parser-1.0
+```
+
+### 3) Run (input + output on the same mounted folder)
+Windows PowerShell:
+```powershell
+docker run --rm --env-file .env `
+  -v "C:\path\to\pdfs:/data" `
+  zev94/radiant-llm:visual-parser-1.0 `
+  --input-dir /data --output-dir /data
+```
+
+Linux / WSL:
+```bash
+docker run --rm --env-file .env \
+  -v "/path/to/pdfs:/data" \
+  zev94/radiant-llm:visual-parser-1.0 \
+  --input-dir /data --output-dir /data
+```
+
+### 4) Run (separate output directory)
+Windows PowerShell:
+```powershell
+docker run --rm --env-file .env `
+  -v "C:\path\to\pdfs:/data" `
+  -v "C:\path\to\out:/out" `
+  zev94/radiant-llm:visual-parser-1.0 `
+  --input-dir /data --output-dir /out
+```
+
+### Offline install (legacy `.tar`)
+
+```powershell
+docker load -i .\visual-parser_0.1.0.tar
+docker images   # use the tag printed by Docker
+```
+
+### Model overrides (optional)
+
+Default vision model is **GPT-5.5** when using `--vision-provider gpt`. Override on the command line:
+
+```powershell
+docker run --rm --env-file .env -v "C:\path\to\pdfs:/data" `
+  zev94/radiant-llm:visual-parser-1.0 `
+  --input-dir /data --output-dir /data --vision-model gpt-5.4
+```
+
+<!-- ## Run from source (Python)
+
+From `codebase/Visual-Parser/`:
+```powershell
+python visual-parser.py --input-dir "C:\path\to\pdfs"
+``` -->
+
+## Common configuration flags
+
+After pulling the image, run:
+
+```bash
+docker run --rm zev94/radiant-llm:visual-parser-1.0 --help
+```
+
+For copy-paste **Docker** examples (vision presets, text modes, workers, rebuild), see [`docker-usage-examples.md`](docker-usage-examples.md).
+
+Paths:
+- `--input-dir` / `-i` (required)
+- `--output-dir` / `-o` (default: same as input)
+
+Text extraction:
+- `--text-mode nougat|lightweight` (default: `nougat`)
+- `--nougat-model facebook/nougat-small`
+- `--chunk-size 500`
+- `--chunk-overlap 100`
+
+Vision LLM:
+- `--vision-provider gpt|gemini` (default: `gpt`)
+- `--vision-model gpt-5.2` (or `gpt-4o`, `gemini-2.5-flash`, etc.)
+- `--vision-detail low|high|auto`
+- `--reasoning-effort none|low|medium|high|xhigh`
+- `--metadata-pages 2`
+
+Performance / misc:
+- `--max-workers 4`
+- `--rebuild` (reprocess everything; ignore `04_processed_pdfs.txt`)
+- `--log-level DEBUG|INFO|WARNING|ERROR`
+
+---
+
+## Citation
+
+If you use RADIANT-LLM or the accompanying evaluation materials, please cite the preprint:
+
+```bibtex
+@article{ndum2026radiant,
+  title={RADIANT-LLM: an Agentic Retrieval Augmented Generation Framework for Reliable Decision Support in Safety-Critical Nuclear Engineering},
+  author={Ndum, Zavier Ndum and Tao, Jian and Ford, John and Yim, Mansung and Liu, Yang},
+  journal={arXiv preprint arXiv:2604.22755},
+  year={2026}
+}
+```
+
+Preprint: https://arxiv.org/abs/2604.22755
+
+---
+
+## License
+
+This repository is currently proprietary and not licensed for public use, redistribution, or modification. Licensing terms will be updated after institutional review.
+
+

visual_parser-1.0.0/pyproject.toml

@@ -0,0 +1,67 @@
+[build-system]
+requires      = ["setuptools>=61", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name        = "visual-parser"
+version     = "1.0.0"
+description = "Standalone Visual-RAG PDF Parser — text extraction + Vision-LLM figure descriptions → JSONL"
+readme      = "README.md"
+requires-python = ">=3.10"
+license     = { text = "MIT" }
+
+keywords = [
+    "pdf", "rag", "nougat", "vision-llm", "ocr",
+    "document-parsing", "jsonl", "knowledge-base",
+]
+
+classifiers = [
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "License :: OSI Approved :: MIT License",
+    "Operating System :: OS Independent",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Topic :: Text Processing :: Markup",
+    "Intended Audience :: Science/Research",
+]
+
+dependencies = [
+    "PyMuPDF==1.24.7",
+    "Pillow==11.1.0",
+    "torch==2.7.0",
+    "transformers==4.45.2",
+    "huggingface-hub==0.36.0",
+    "langchain-community==0.3.3",
+    "langchain==0.3.13",
+    "langchain-text-splitters==0.3.4",
+    "openai==1.78.1",
+    "google-generativeai==0.8.5",
+    "python-dotenv==1.1.0",
+    "tqdm==4.67.1",
+]
+
+[project.optional-dependencies]
+ocr = ["pytesseract==0.3.13"]   # requires Tesseract binary installed separately
+dev = ["pytest", "ruff", "mypy"]
+
+[project.urls]
+Homepage      = "https://github.com/SmartLabNuclear/RADIANT_LLM"
+Repository    = "https://github.com/SmartLabNuclear/RADIANT_LLM"
+"Docker Hub"  = "https://hub.docker.com/r/zev94/radiant-llm"
+
+[project.scripts]
+visual-parser = "visual_parser.cli_main:main"
+
+[tool.setuptools.packages.find]
+where = ["."]
+include = ["visual_parser*"]
+
+[tool.ruff]
+line-length = 100
+target-version = "py310"
+
+[tool.mypy]
+python_version = "3.10"
+ignore_missing_imports = true

visual_parser-1.0.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

visual_parser-1.0.0/visual_parser/__init__.py

@@ -0,0 +1,20 @@
+"""
+visual_parser — Standalone Visual-RAG PDF Parser
+=================================================
+Detects new PDFs in a user-supplied directory, extracts text (via Nougat or
+lightweight PyMuPDF/PyPDFLoader), describes every figure/chart/schematic using
+a Vision LLM (OpenAI GPT-4o or Google Gemini), and writes three JSONL knowledge
+bases ready for any downstream RAG system:
+
+    01_chunks_kb.jsonl   – text chunks with stable IDs
+    02_visuals_kb.jsonl  – per-figure visual descriptions
+    03_metadata_kb.jsonl – document-level metadata (title, authors, DOI …)
+
+No chatbot, no vector store, no retrieval – just a robust parser.
+"""
+
+from visual_parser.config import ParserConfig
+from visual_parser.pipeline import run_pipeline
+
+__all__ = ["ParserConfig", "run_pipeline"]
+__version__ = "1.0.0"

visual_parser-1.0.0/visual_parser/__main__.py

@@ -0,0 +1,8 @@
+"""
+Enables:  python -m visual_parser --input-dir ./my_pdfs ...
+"""
+import sys
+from visual_parser.cli_main import main
+
+if __name__ == "__main__":
+    sys.exit(main())

visual_parser-1.0.0/visual_parser/cli.py

@@ -0,0 +1,230 @@
+"""
+cli.py — Argument parser and main() entry point for the Visual-RAG PDF Parser.
+
+This module is the canonical home for CLI logic.  It is imported by:
+    • visual-parser.py          (top-level convenience script)
+    • visual_parser/__main__.py (enables: python -m visual_parser ...)
+    • pyproject.toml [project.scripts] (enables: visual-parser ...)
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+import sys
+
+
+USAGE_EXAMPLES = """
+Examples
+--------
+# Nougat (default) + GPT-5.5 vision
+python visual-parser.py --input-dir ./my_pdfs
+
+# Fast lightweight extraction + Gemini
+python visual-parser.py --input-dir ./my_pdfs \\
+    --text-mode lightweight \\
+    --vision-provider gemini \\
+    --vision-model gemini-1.5-pro
+
+# Write outputs to a separate directory
+python visual-parser.py --input-dir ./my_pdfs --output-dir ./output_kb
+
+# Force re-parse all PDFs (ignore tracking registry)
+python visual-parser.py --input-dir ./my_pdfs --rebuild
+
+# High-detail images for dense schematics
+python visual-parser.py --input-dir ./my_pdfs --vision-detail high
+
+# Verbose console logging
+python visual-parser.py --input-dir ./my_pdfs --log-level INFO
+"""
+
+
+def _build_arg_parser() -> argparse.ArgumentParser:
+    p = argparse.ArgumentParser(
+        prog="visual-parser",
+        description=(
+            "Visual-RAG PDF Parser — detects new PDFs, extracts text and "
+            "figure descriptions, and writes three JSONL knowledge bases:\n"
+            "  01_chunks_kb.jsonl   text chunks\n"
+            "  02_visuals_kb.jsonl  visual descriptions\n"
+            "  03_metadata_kb.jsonl document metadata"
+        ),
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog=USAGE_EXAMPLES,
+    )
+
+    # ---- Paths --------------------------------------------------------------
+    io_group = p.add_argument_group("Paths")
+    io_group.add_argument(
+        "--input-dir", "-i",
+        required=True,
+        metavar="DIR",
+        help="Directory to scan for PDF files (searched recursively).",
+    )
+    io_group.add_argument(
+        "--output-dir", "-o",
+        default="",
+        metavar="DIR",
+        help="Directory where JSONL files are written. Defaults to --input-dir.",
+    )
+
+    # ---- Text extraction ----------------------------------------------------
+    text_group = p.add_argument_group("Text extraction")
+    text_group.add_argument(
+        "--text-mode",
+        choices=["nougat", "lightweight"],
+        default="nougat",
+        help=(
+            "nougat      — Nougat OCR model (best for scanned/complex PDFs, GPU recommended).\n"
+            "lightweight — PyMuPDF text layer + PyPDFLoader fallback (fast, no GPU needed)."
+        ),
+    )
+    text_group.add_argument(
+        "--nougat-model",
+        default="facebook/nougat-small",
+        metavar="MODEL_ID",
+        help="HuggingFace model ID for Nougat (default: facebook/nougat-small).",
+    )
+    text_group.add_argument(
+        "--chunk-size",
+        type=int,
+        default=500,
+        metavar="N",
+        help="Target characters per text chunk (default: 500).",
+    )
+    text_group.add_argument(
+        "--chunk-overlap",
+        type=int,
+        default=100,
+        metavar="N",
+        help="Overlap characters between adjacent chunks (default: 100).",
+    )
+
+    # ---- Vision LLM ---------------------------------------------------------
+    vision_group = p.add_argument_group("Vision LLM (figure descriptions & metadata)")
+    vision_group.add_argument(
+        "--vision-provider",
+        choices=["gpt", "gemini"],
+        default="gpt",
+        help=(
+            "gpt    — OpenAI GPT-5.5  (set OPENAI_API_KEY in .env).\n"
+            "gemini — Google Gemini   (set GEMINI_API_KEY in .env)."
+        ),
+    )
+    vision_group.add_argument(
+        "--vision-model",
+        default=None,
+        metavar="MODEL_NAME",
+        help=(
+            "Vision model name. Omit to use the latest for each provider:\n"
+            "  gpt    → gpt-5.5            (also: gpt-5.4, gpt-5.3-chat-latest, gpt-5.2, gpt-5.1, gpt-5, gpt-4o, gpt-4.1)\n"
+            "  gemini → gemini-3-pro-preview (also: gemini-2.5-flash, gemini-1.5-pro)"
+        ),
+    )
+    vision_group.add_argument(
+        "--vision-detail",
+        choices=["low", "high", "auto"],
+        default="low",
+        help=(
+            "Image detail level (GPT only).\n"
+            "low  — faster/cheaper (default, recommended for most use cases).\n"
+            "high — better for dense schematics with small text."
+        ),
+    )
+    vision_group.add_argument(
+        "--reasoning-effort",
+        choices=["minimal", "none", "low", "medium", "high", "xhigh"],
+        default="medium",
+        help=(
+            "Reasoning effort for GPT-5.x models (ignored for Gemini and older GPT).\n"
+            "  minimal/none — minimum reasoning, depending on model.\n"
+            "  low          — light reasoning.\n"
+            "  medium       — balanced (default).\n"
+            "  high         — deeper reasoning, slower.\n"
+            "  xhigh        — maximum depth (gpt-5.2, gpt-5.4, and gpt-5.5)."
+        ),
+    )
+    vision_group.add_argument(
+        "--metadata-pages",
+        type=int,
+        default=2,
+        metavar="N",
+        help="Number of front pages sent to the vision LLM for metadata extraction (default: 2).",
+    )
+
+    # ---- Performance --------------------------------------------------------
+    perf_group = p.add_argument_group("Performance")
+    perf_group.add_argument(
+        "--max-workers",
+        type=int,
+        default=4,
+        metavar="N",
+        help="Thread-pool size for parallel PDF processing (default: 4).",
+    )
+
+    # ---- Misc ---------------------------------------------------------------
+    misc_group = p.add_argument_group("Miscellaneous")
+    misc_group.add_argument(
+        "--rebuild",
+        action="store_true",
+        help=(
+            "Reprocess ALL PDFs, ignoring the 04_processed_pdfs.txt registry. "
+            "Use after changing prompts, chunking strategy, or switching models."
+        ),
+    )
+    misc_group.add_argument(
+        "--log-level",
+        choices=["DEBUG", "INFO", "WARNING", "ERROR"],
+        default="ERROR",
+        help="Verbosity level written to 05_pipeline.log (default: ERROR).",
+    )
+
+    return p
+
+
+def main(argv=None) -> int:
+    """
+    Parse CLI arguments, build a :class:`~visual_parser.config.ParserConfig`,
+    validate it, and hand off to :func:`~visual_parser.pipeline.run_pipeline`.
+
+    Returns 0 on success, 1 on configuration error.
+    """
+    parser = _build_arg_parser()
+    args   = parser.parse_args(argv)
+
+    # Default vision model per provider when not explicitly set
+    if args.vision_model is None:
+        args.vision_model = (
+            "gpt-5.5" if args.vision_provider == "gpt" else "gemini-3-pro-preview"
+        )
+
+    from visual_parser.config import ParserConfig
+
+    config = ParserConfig(
+        input_dir             = os.path.abspath(args.input_dir),
+        output_dir            = os.path.abspath(args.output_dir) if args.output_dir else "",
+        text_mode             = args.text_mode,
+        nougat_model          = args.nougat_model,
+        chunk_size            = args.chunk_size,
+        chunk_overlap         = args.chunk_overlap,
+        vision_provider       = args.vision_provider,
+        gpt_vision_model      = args.vision_model if args.vision_provider == "gpt"    else "gpt-5.5",
+        gemini_vision_model   = args.vision_model if args.vision_provider == "gemini" else "gemini-3-pro-preview",
+        gpt_reasoning_effort  = args.reasoning_effort,
+        vision_detail         = args.vision_detail,
+        metadata_pages        = args.metadata_pages,
+        max_workers           = args.max_workers,
+        rebuild               = args.rebuild,
+        log_level             = args.log_level,
+    )
+
+    try:
+        config.validate()
+    except ValueError as exc:
+        print(f"[ERROR] {exc}", file=sys.stderr)
+        return 1
+
+    from visual_parser.pipeline import run_pipeline
+    run_pipeline(config)
+    return 0