ingestlib 0.1.0__tar.gz

ingestlib-0.1.0/.claude/settings.local.json

@@ -0,0 +1,11 @@
+{
+  "permissions": {
+    "allow": [
+      "WebFetch(domain:www.llamaindex.ai)",
+      "WebFetch(domain:developers.llamaindex.ai)",
+      "WebSearch",
+      "WebFetch(domain:aws.amazon.com)",
+      "Bash(uv run *)"
+    ]
+  }
+}

ingestlib-0.1.0/.env.example

@@ -0,0 +1,9 @@
+# Jina AI Reranker — https://jina.ai/api-dashboard/
+JINA_API_KEY=jina_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+
+# Pinecone — https://app.pinecone.io/ → API Keys
+PINECONE_API_KEY=pcsk_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+
+# Qdrant — https://cloud.qdrant.io/ → API Keys, or a self-hosted server
+QDRANT_URL=http://localhost:6333
+QDRANT_API_KEY=

ingestlib-0.1.0/.gitignore

@@ -0,0 +1,230 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#   Usually these files are written by a python script from a template
+#   before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+# Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+# uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+# poetry.lock
+# poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+# pdm.lock
+# pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+# pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# Redis
+*.rdb
+*.aof
+*.pid
+
+# RabbitMQ
+mnesia/
+rabbitmq/
+rabbitmq-data/
+
+# ActiveMQ
+activemq-data/
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#   JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#   be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#   and can be added to the global gitignore or merged into this file.  For a more nuclear
+#   option (not recommended) you can uncomment the following to ignore the entire idea folder.
+# .idea/
+
+# Abstra
+#   Abstra is an AI-powered process automation framework.
+#   Ignore directories containing user credentials, local state, and settings.
+#   Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#   Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
+#   that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#   and can be added to the global gitignore or merged into this file. However, if you prefer, 
+#   you could uncomment the following to ignore the entire vscode folder
+# .vscode/
+# Temporary file for partial code execution
+tempCodeRunnerFile.py
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/
+
+# Streamlit
+.streamlit/secrets.toml
+
+# macOS
+*.DS_Store
+
+# Internal design notes — not for the public repo
+plan.md
+
+# Personal settings — copy config.example.yaml and fill in your values
+config.yaml
+
+# Eval score snapshots — local measurement history
+evals/results/

ingestlib-0.1.0/.python-version

@@ -0,0 +1 @@
+3.13

ingestlib-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 LangModule
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

ingestlib-0.1.0/Makefile

@@ -0,0 +1,77 @@
+.PHONY: test test-all test-llm test-nova test-embedding test-rerank test-rerank-aws test-rerank-jina test-ocr test-parse test-classify test-split test-s3 test-pinecone test-qdrant test-services eval
+
+# fast suite — every opt-in e2e group skips (RUN_* gates unset)
+test:
+	uv run pytest tests/
+
+# entire suite including every opt-in group (needs VL server running + Bedrock access)
+test-all:
+	RUN_AWS_RERANK=1 RUN_OCR_E2E=1 RUN_PARSE_E2E=1 RUN_CLASSIFY_E2E=1 RUN_SPLIT_E2E=1 RUN_S3_E2E=1 RUN_PINECONE_E2E=1 RUN_QDRANT_E2E=1 RUN_SERVICES_E2E=1 uv run pytest tests/
+
+# --- llm layer (mirrors src/ingestlib/foundations/llm/) ---
+
+test-llm:
+	uv run pytest tests/foundations/llm/
+
+test-nova:
+	uv run pytest tests/foundations/llm/bedrock/nova/
+
+test-embedding:
+	uv run pytest tests/foundations/llm/bedrock/embedding/
+
+# both providers — AWS-hitting tests skip without RUN_AWS_RERANK=1
+test-rerank:
+	uv run pytest tests/foundations/llm/bedrock/rerank/ tests/foundations/llm/jina/
+
+# opt-in: hits amazon.rerank-v1:0 (2 RPM quota — expect ~1 min of throttle sleeps)
+test-rerank-aws:
+	RUN_AWS_RERANK=1 uv run pytest tests/foundations/llm/bedrock/rerank/
+
+test-rerank-jina:
+	uv run pytest tests/foundations/llm/jina/
+
+# --- ocr layer (mirrors src/ingestlib/foundations/ocr/) — needs the VL inference server ---
+
+test-ocr:
+	RUN_OCR_E2E=1 uv run pytest tests/foundations/ocr/
+
+# --- parse operation (mirrors src/ingestlib/operations/parse/) — needs VL server + Bedrock ---
+
+test-parse:
+	RUN_PARSE_E2E=1 uv run pytest tests/operations/parse/
+
+# --- classify operation (mirrors src/ingestlib/operations/classify/) — needs Bedrock ---
+
+test-classify:
+	RUN_CLASSIFY_E2E=1 uv run pytest tests/operations/classify/
+
+# --- split operation (mirrors src/ingestlib/operations/split/) — needs Bedrock ---
+
+test-split:
+	RUN_SPLIT_E2E=1 uv run pytest tests/operations/split/
+
+# --- storage (S3 artifacts) — needs AWS credentials ---
+
+test-s3:
+	RUN_S3_E2E=1 uv run pytest tests/storage/
+
+# --- pinecone connector — needs PINECONE_API_KEY + Bedrock (embeddings) ---
+
+test-pinecone:
+	RUN_PINECONE_E2E=1 uv run pytest tests/storage/pinecone/
+
+# --- qdrant connector — needs a Qdrant server at QDRANT_URL + Bedrock ---
+
+test-qdrant:
+	RUN_QDRANT_E2E=1 uv run pytest tests/storage/qdrant/
+
+# --- services (ingest + retrieve) — needs the FULL stack ---
+
+test-services:
+	RUN_SERVICES_E2E=1 uv run pytest tests/services/
+
+# --- retrieval quality eval — measurement, not a test (needs the full stack;
+#     first run also needs the VL server to ingest the fixture corpus) ---
+
+eval:
+	uv run python evals/run_eval.py

ingestlib-0.1.0/PKG-INFO

@@ -0,0 +1,248 @@
+Metadata-Version: 2.4
+Name: ingestlib
+Version: 0.1.0
+Summary: Self-hosted document intelligence for RAG — parse, classify, and split PDF/DOCX/PPTX into searchable, cited, retrieval-ready chunks
+Project-URL: Homepage, https://github.com/LangModule/ingestlib
+Project-URL: Repository, https://github.com/LangModule/ingestlib
+Project-URL: Issues, https://github.com/LangModule/ingestlib/issues
+Author-email: ramacharanreddy-k <ramacharanreddykasireddy@gmail.com>
+License-Expression: MIT
+License-File: LICENSE
+Keywords: chunking,document-intelligence,document-parsing,hybrid-search,ocr,rag,retrieval
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: MacOS
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Text Processing :: Indexing
+Requires-Python: >=3.12
+Requires-Dist: boto3>=1.43.29
+Requires-Dist: botocore[crt]>=1.43
+Requires-Dist: fastembed>=0.8.0
+Requires-Dist: httpx>=0.27
+Requires-Dist: langchain-aws>=1.5.1
+Requires-Dist: mlx-vlm>=0.6.4; sys_platform == 'darwin' and platform_machine == 'arm64'
+Requires-Dist: paddleocr>=3.7.0
+Requires-Dist: paddlepaddle>=3.3.1
+Requires-Dist: paddlex[genai-client,ocr]==3.7.1
+Requires-Dist: pillow>=10.0
+Requires-Dist: pinecone>=9.1.0
+Requires-Dist: pydantic>=2.0
+Requires-Dist: pypdfium2>=4.0
+Requires-Dist: python-dotenv>=1.0
+Requires-Dist: pyyaml>=6.0.2
+Requires-Dist: qdrant-client>=1.18.0
+Description-Content-Type: text/markdown
+
+# ingestlib
+
+![ingestlib — self-hosted document intelligence for RAG](assets/cover.png)
+
+Self-hosted document intelligence for RAG pipelines. One library that takes a
+raw document — PDF, DOCX, PPTX — and produces searchable, **cited**,
+retrieval-ready chunks: the territory of LlamaParse / Reducto /
+Unstructured.io, running on your own stack.
+
+```python
+from ingestlib.services import ingest, retrieve
+
+ingest("finance-10k.pdf")            # parse → classify → split → embed → vector store
+result = retrieve("what were the total revenues?")
+print(result.context)                # ranked chunks, each citing doc · page · section
+```
+
+## What it does
+
+| Stage | What you get |
+|---|---|
+| **Parse** | Layout-aware markdown per page: tables as HTML (merged cells intact), formulas as LaTeX, **charts converted to data tables** (estimated values marked `~`, printed callouts and growth labels captured), figures extracted as PNG crops with captions and AI descriptions — every block traceable to a bounding box on the page |
+| **Classify** | Document-type label (`invoice`, `research_paper`, …) — open-ended or constrained to your categories, with confidence and alternatives. Works standalone with **no OCR** |
+| **Split** | Sections (pages grouped by role: `methods`, `results`, …) containing **natural chunks** — boundaries follow the content, tables never split, each chunk carries a `[category › section › heading]` breadcrumb in its `embedding_text` |
+| **Ingest** | The whole pipeline in one call, every stage persisted to S3, vectors upserted, deduplicated by content checksum |
+| **Retrieve** | Question → **hybrid search** (dense embeddings + lexical sparse, merged) → **Jina rerank** → hits with scores and citations, plus a prompt-ready context block |
+
+Engines: **PaddleOCR-VL-1.6** (0.9B VLM, runs on your GPU) for layout + recognition,
+**Amazon Nova 2 Lite** for judgment (chart reading, review, classification,
+chunk boundaries), **Nova multimodal embeddings**, **Pinecone or Qdrant** for
+vectors (both hybrid dense + sparse), **S3** for artifacts. ~$0.002/page in LLM spend.
+
+## Quickstart
+
+### 1. Requirements
+
+- Python 3.12+ and [uv](https://github.com/astral-sh/uv)
+- **AWS account** with Bedrock access (`us-east-1`): Nova 2 Lite + Nova 2
+  multimodal embeddings
+- **Vector database** — Pinecone account (serverless, free tier works;
+  the default) or a Qdrant server (local docker or Qdrant Cloud)
+- **Jina AI account** for reranking (free tier: 100 RPM)
+
+### 2. Install
+
+```bash
+git clone https://github.com/LangModule/ingestlib.git
+cd ingestlib
+uv sync
+```
+
+System dependency — LibreOffice (DOCX/PPTX → PDF conversion):
+
+```bash
+brew install --cask libreoffice          # macOS (binary is `soffice`)
+sudo apt install libreoffice-core libreoffice-writer libreoffice-impress   # Linux
+```
+
+### 3. Start the OCR inference server
+
+Parse runs PaddleOCR-VL-1.6 behind an inference server. First launch downloads
+~1.8 GB of weights; later launches load from cache in seconds.
+
+```bash
+# Apple Silicon (Metal GPU)
+uv run python -m mlx_vlm.server --port 8111 --model PaddlePaddle/PaddleOCR-VL-1.6
+
+# NVIDIA (then set paddle_vl.backend: vllm-server in config.yaml)
+vllm serve PaddlePaddle/PaddleOCR-VL-1.6 --port 8111
+```
+
+The layout model (PP-DocLayoutV3, ~126 MB) auto-downloads on the first parse.
+
+### 4. Configure
+
+```bash
+cp .env.example .env                 # paste your API keys (Jina + Pinecone and/or Qdrant)
+cp config.example.yaml config.yaml   # your AWS profile, bucket name, vector store choice
+aws configure --profile your-aws-profile   # Bedrock-enabled credentials
+```
+
+Edit `config.yaml`: your AWS profile/account, S3 bucket name (globally
+unique), vector store choice and index/collection names. **The S3 bucket and
+the vector indexes/collection are created automatically on first use** — no
+manual setup.
+
+Config is discovered at call time, never at import: `INGESTLIB_CONFIG=/path/to/config.yaml`
+wins, otherwise the working directory and its parents are searched — so
+installed usage works the same as running inside this repo.
+
+### 5. Run
+
+```python
+from ingestlib.services import ingest, retrieve
+
+r = ingest("report.pdf")
+print(r.status, r.category, r.chunks, r.durations)
+
+res = retrieve("what does the report conclude?", top_k=5)
+for hit in res.hits:
+    print(hit.rerank_score, hit.citation, hit.chunk.heading)
+```
+
+## Using the operations directly
+
+Every operation also works standalone:
+
+```python
+from ingestlib.operations import parse, classify, split
+
+result = parse("report.pdf")            # ParseResult: pages, regions, figures
+print(result.markdown)                  # whole-document markdown
+result.save_images("out/")              # extracted figures/charts as PNGs
+
+label = classify("report.pdf")          # no OCR needed — native text + embedded images
+chunks = split(result, category=label.category)
+for c in chunks.chunks:
+    print(c.token_estimate, c.embedding_text.splitlines()[0])
+```
+
+Persistence and vector access are explicit too:
+
+```python
+from ingestlib.storage import artifacts, PineconeStore
+
+doc_id = artifacts.save_parse(result)   # S3: source, result.json, page PNGs, crops
+artifacts.list_documents()              # registry: filename, pages, category, chunks
+```
+
+## Architecture
+
+```
+src/ingestlib/
+├── services/       ingest · retrieve          — the product
+├── operations/     parse · classify · split   — the tools (each standalone)
+├── storage/        artifacts (S3) · base (VectorStore contract) · pinecone · qdrant
+├── foundations/    llm (Bedrock Nova, Jina) · ocr (PaddleOCR-VL)
+├── utils/          logger · files
+└── config.py       config.yaml + .env → typed configs
+```
+
+Strict downward dependencies. The `VectorStore` contract means backends drop
+in as connectors — both ship **hybrid search**: **Pinecone** (dense + hosted
+sparse model, merged client-side) and **Qdrant** (dense + BM25 with
+server-side IDF and RRF fusion; local docker or cloud). Pick one with
+`vector_store: pinecone | qdrant` in config.yaml. Keys for both can sit in
+`.env` — only the selected connector ever builds a client.
+
+## Logging
+
+```bash
+INGESTLIB_LOG_LEVEL=INFO           # DEBUG | INFO | WARNING | ERROR (default INFO)
+INGESTLIB_LOG_THIRD_PARTY=1        # also show paddlex/httpx/botocore chatter
+INGESTLIB_LOG_COLOR=0              # disable colored output
+```
+
+## Testing
+
+Tests hit **real APIs, never mocks**. Pure logic runs always; server-hitting
+suites are opt-in via env gates.
+
+```bash
+make test                  # fast suite (~180 tests, ~90s; e2e groups skip)
+make test-parse            # parse e2e            (needs VL server + Bedrock)
+make test-classify         # classify e2e         (needs Bedrock)
+make test-split            # split e2e            (needs Bedrock)
+make test-s3               # artifact store e2e   (needs AWS)
+make test-pinecone         # vector connector e2e (needs Pinecone + Bedrock)
+make test-qdrant           # vector connector e2e (needs a Qdrant server + Bedrock)
+make test-services         # full product e2e     (needs the entire stack)
+make test-all              # everything
+make eval                  # retrieval quality eval (see below)
+```
+
+Fixture PDFs live in `tests/data/pdf/` — 14 real documents (research papers,
+earnings decks, insurance forms, timetables, 10-Ks).
+
+### Retrieval quality
+
+Beyond pass/fail tests, `evals/` measures retrieval quality: 22 ground-truth
+questions over the fixture corpus, run through the real `retrieve()` flow
+under dense/hybrid × rerank on/off, scored by hit@k and MRR. Measured so far:
+**with reranking, every answer lands in the top 3 hits (hit@3 = 1.00)**;
+hit@1 ranges 0.86–1.00 across runs. Each run saves a timestamped snapshot to
+`evals/results/`, so quality changes are visible over time.
+
+## Disk footprint
+
+| Component | Size | Location |
+|---|---|---|
+| Python deps | ~3 GB | `.venv/` |
+| PaddleOCR-VL-1.6 weights | ~1.8 GB | `~/.cache/huggingface/hub/` |
+| PP-DocLayoutV3 | ~126 MB | `~/.paddlex/official_models/` |
+| LibreOffice | ~600 MB | system |
+
+## Scope
+
+English documents; PDF / DOCX / PPTX input. Images, charts, and tables
+**inside** documents are fully extracted and interpreted; direct image files
+and handwriting are out of scope by design.
+
+## Roadmap
+
+- Hover-highlight review UI (bbox provenance already shipped for it)
+- Extract: schema-driven field extraction with source provenance
+
+## License
+
+See [LICENSE](./LICENSE).