SemantixRAG — AI-Native Data Platform (v2.0)

A production-grade, open-source AI-native data platform featuring end-to-end RAG ingestion, knowledge graph integration (GraphRAG), AI observability (Obsidian), automated compliance (GuardRail), multi-modal extraction, and a REST API server — all running locally with zero cloud dependencies.

🌐 Website: semantixrag.github.io
📦 GitHub: github.com/SemantixRAG/SemantixRAG
📖 License: MIT

---

Platform Overview

Core Components

Layer	Components	Features
API & Auth	FastAPI + OPA	REST API, Policy Engine, Rate Limiting
Ingestion Pipeline	Extraction → Chunking → Embedding → Indexing	Multi-modal, VLM, Whisper, OCR
Storage	OpenSearch + Neo4j	Vector DB + Knowledge Graph
Features	Obsidian, GuardRail, GraphRAG, AdminCopilot	Tracing, PII Detection, Entity Extraction, Cost Tracking

Ingestion Flow

Document → VLM/Whisper (multimodal) 
         → Chunking (header-aware splitting)
         → Enrichment (summarization + NER)
         → Embedding (BAAI/bge-m3 + JinaCLIP)
         → Indexing (OpenSearch k-NN + BM25)
         → GraphRAG (Neo4j entity linking)
         → PII Masking (GuardRail)

Observability & Compliance

Subsystem	Functions
Obsidian (Tracing)	Distributed tracing, RAG quality metrics, cost attribution
GuardRail (Compliance)	PII detection/masking, GDPR DSAR, audit logging
GraphRAG (Knowledge)	NER, entity linking, multi-hop traversal, ontologies
AdminCopilot (Admin)	Natural-language platform administration

Technology Stack

Component	Technology	Status
Orchestration	Python 3.11+	Core
Containerization	Docker & Docker Compose	Core
Vector Database	OpenSearch (k-NN + BM25 search)	Core
Knowledge Graph	Neo4j 5.15 (Cypher + APOC)	NEW
API Server	FastAPI + Uvicorn + Gunicorn	NEW
Policy Engine	Open Policy Agent (OPA) + Rego	NEW
PII Detection	Microsoft Presidio + Regex fallback	NEW
Parsing	Unstructured.io, PyMuPDF, PyPDF	Core
VLM Fallback	LLaVA / ColPali (images)	Core
Audio Transcription	OpenAI Whisper	NEW
Embeddings	BAAI/bge-m3 (1024-dim, multilingual)	Core
Multi-modal Embeddings	JinaCLIP (text+image), CLAP (audio)	NEW
Summarization	Gemma / Llama (via Ollama or HuggingFace)	Core
CDC	Python Watchdog	Core
Testing	Pytest + pytest-asyncio + pytest-cov (41+ tests)	NEW
CI/CD	GitHub Actions (lint, test, Trivy, Docker)	NEW

Project Structure (v2.0)

SemantixRAG/
├── .github/workflows/
│   └── ci.yml                      # CI/CD pipeline
├── config/
│   ├── __init__.py
│   ├── settings.py                 # Pydantic settings (enhanced)
│   └── opa/
│       ├── access.rego             # RBAC/ABAC policy
│       ├── masking.rego            # Conditional masking policy
│       └── audit.rego              # Audit level classification
├── docker/
│   ├── docker-compose.yml          # OpenSearch + Neo4j + Redis + OPA
│   ├── opensearch.yml              # OpenSearch configuration
│   └── Dockerfile                  # Python app container
├── documents/
│   └── sample_document.md
├── src/
│   ├── __init__.py
│   ├── models.py                   # Enhanced Pydantic models
│   ├── pipeline.py                 # Enhanced orchestrator with P0 features
│   ├── api/                        # NEW: FastAPI server
│   │   ├── __init__.py
│   │   ├── main.py                 # FastAPI entry point
│   │   └── routes/
│   │       ├── __init__.py
│   │       ├── ingestion.py        # POST /v1/ingest
│   │       ├── retrieval.py        # POST /v1/query
│   │       ├── compliance.py       # POST /v1/compliance/pii/scan, /dsar
│   │       ├── observability.py    # POST /v1/observability/traces
│   │       └── admin.py            # POST /v1/admin/query (AdminCopilot)
│   ├── compliance/                 # NEW: GuardRail
│   │   ├── __init__.py
│   │   ├── pii_scanner.py          # Presidio + regex PII detection
│   │   ├── masking.py              # Dynamic PII masking engine
│   │   └── dsar.py                 # GDPR DSAR automation
│   ├── knowledge/                  # NEW: GraphRAG
│   │   ├── __init__.py
│   │   ├── entity_extractor.py     # spaCy NER + entity linking
│   │   └── ontology.py             # Domain ontologies + auto-discovery
│   ├── observability/              # NEW: Obsidian
│   │   ├── __init__.py
│   │   ├── tracer.py               # Distributed tracing
│   │   ├── evaluator.py            # RAG quality metrics
│   │   └── metrics.py              # Counters, histograms, cost tracking
│   ├── extractors/
│   │   ├── __init__.py
│   │   ├── base.py
│   │   ├── unstructured_extractor.py
│   │   ├── table_extractor.py
│   │   └── multimodal_extractor.py # NEW: VLM + Whisper + video
│   ├── chunking/
│   │   ├── __init__.py
│   │   ├── header_splitter.py
│   │   └── enricher.py
│   ├── embeddings/
│   │   ├── __init__.py
│   │   └── embedder.py
│   ├── indexing/
│   │   ├── __init__.py
│   │   ├── connection.py
│   │   ├── index_manager.py
│   │   ├── bulk_indexer.py
│   │   ├── hybrid_search.py
│   │   └── graph_writer.py         # NEW: Neo4j async writer
│   ├── cdc/
│   │   ├── __init__.py
│   │   ├── watcher.py
│   │   └── incremental.py
│   └── monitoring/
│       ├── __init__.py
│       └── logger.py
├── tests/
│   ├── test_knowledge.py           # NEW: 11 GraphRAG tests
│   ├── test_compliance.py          # NEW: 12 GuardRail tests
│   └── test_observability.py       # NEW: 18 Obsidian tests
├── main.py                         # CLI entry point
├── requirements.txt                # Enhanced dependencies
├── .env.example                    # Enhanced config template
├── README.md
└── index.html                      # Project landing page (v2.0)

Quick Start

1. Clone & Setup

git clone https://github.com/SemantixRAG/SemantixRAG.git
cd SemantixRAG

2. Start Infrastructure

cd docker
docker-compose up -d

This starts all platform services:

• OpenSearch on localhost:9200 (vector store + BM25 search)
• OpenSearch Dashboards on localhost:5601 (visualization)
• Neo4j on localhost:7687 / 7474 (knowledge graph)
• Redis on localhost:6379 (caching + queue)
• OPA on localhost:8181 (policy engine)

3. Install Python Dependencies

python -m venv .venv
.venv\Scripts\activate    # Windows
# source .venv/bin/activate  # Linux/Mac

pip install -r requirements.txt

4. Initialize Indexes

python main.py init

5. Ingest Documents

python main.py ingest ./documents/sample_document.md
python main.py ingest ./documents/

6. Search

python main.py search "machine learning"
python main.py search "reinforcement learning" --top-k 10

7. Start API Server

python -m uvicorn src.api.main:app --reload

Now you can query via REST:

curl -X POST http://localhost:8000/v1/query \
  -H "Content-Type: application/json" \
  -d '{"query": "reinforcement learning", "strategy": "graph"}'

8. Run Tests

python -m pytest tests/ -v

CLI Commands

Command	Description
init	Initialize OpenSearch index with k-NN mapping
ingest <path>	Ingest a document or directory with entity extraction + PII scan
watch <dir>	Watch a directory for file changes (CDC) with auto-reindex
search <query>	Search indexed documents with hybrid retrieval
stats	Show index statistics

API Endpoints

Method	Endpoint	Description
GET	/health	Health check
POST	/v1/ingest	Upload and process a document
GET	/v1/ingest/{id}/status	Check ingestion status
POST	/v1/query	Semantic search with hybrid/graph retrieval
POST	/v1/admin/query	Natural-language platform administration
POST	/v1/observability/traces	Ingest telemetry traces
GET	/v1/observability/metrics	Query pipeline metrics
GET	/v1/observability/evaluation	Query RAG quality metrics
POST	/v1/compliance/pii/scan	Scan text for PII
POST	/v1/compliance/dsar	Execute GDPR DSAR request
GET	/v1/compliance/dsar/{id}	Check DSAR status

P0 Products (v2.0)

🕵️ Obsidian — AI Observability

• End-to-end distributed tracing for every pipeline stage (extraction, chunking, enrichment, embedding, indexing, graph write, PII scan)
• RAG quality metrics: faithfulness, answer relevancy, context precision, MRR, Recall@k, Precision@k
• Cost tracking with per-operation, per-model, per-tenant attribution
• Latency histograms with P50/P95/P99 aggregation
• Sampling configuration for high-volume pipelines

🕸️ GraphRAG — Knowledge Graph Integration

• Automatic named entity recognition via spaCy at ingestion time
• Entity linking: chunks → entities via MENTIONS relationships in Neo4j
• Entity resolution and coreference grouping
• Multi-hop Cypher traversal: search_related_entities(names, hops=2)
• Domain ontologies: general, healthcare, legal, finance with auto-discovery
• Graph traversal results fused with vector + BM25 via RRF

🛡️ GuardRail — Automated Compliance

• PII detection via Microsoft Presidio (30+ types, score-thresholded) with regex fallback
• Dynamic masking: type-specific tokens ([EMAIL], [SSN]) or uniform masking
• Risk level classification: low / medium / high based on PII type
• GDPR DSAR automation: find / delete / export subject data across document, chunk, embedding, and memory indexes
• OPA policy engine: Rego policies for access control, masking strategy, audit level

🎨 Multi-Modal RAG

• Images: VLM (LLaVA) captioning for images
• Audio: Whisper transcription for MP3, WAV, M4A, OGG, FLAC
• Video: OpenCV frame sampling at configurable intervals
• Text-only fallback when models are unavailable (mock mode)

Configuration

All settings configurable via .env file (prefix RAG_):

Core

Variable	Default	Description
RAG_OPENSEARCH_HOST	localhost	OpenSearch host
RAG_OPENSEARCH_PORT	9200	OpenSearch port
RAG_OPENSEARCH_INDEX	rag_documents	Index name
RAG_EMBEDDING_MODEL_NAME	BAAI/bge-m3	Embedding model
RAG_EMBEDDING_DIMENSION	1024	Vector dimension
RAG_CHUNK_MAX_TOKENS	512	Max tokens per chunk
RAG_CHUNK_OVERLAP_TOKENS	64	Chunk overlap tokens
RAG_WATCH_DIRECTORY	./documents	Watch directory

Neo4j (GraphRAG)

Variable	Default	Description
RAG_NEO4J_URI	bolt://localhost:7687	Neo4j connection URI
RAG_NEO4J_USER	neo4j	Neo4j username
RAG_NEO4J_PASSWORD	password	Neo4j password
RAG_NEO4J_DATABASE	rag	Neo4j database name

Observability (Obsidian)

Variable	Default	Description
RAG_OBSERVABILITY_ENABLED	True	Enable tracing
RAG_OBSERVABILITY_INDEX	rag_observability	OpenSearch telemetry index
RAG_OBSERVABILITY_SAMPLE_RATE	1.0	Trace sampling rate (0-1)

Compliance (GuardRail)

Variable	Default	Description
RAG_PII_SCAN_ENABLED	True	Enable PII scanning on ingestion
RAG_PII_SCAN_DEPTH	standard	Scan depth (standard/deep)
RAG_MASKING_ENABLED	True	Auto-mask PII in chunks
RAG_AUDIT_LOG_ENABLED	True	Enable audit logging

Knowledge Graph

Variable	Default	Description
RAG_ENTITY_EXTRACTION_ENABLED	True	Enable NER at ingestion
RAG_ENTITY_CONFIDENCE_THRESHOLD	0.8	Minimum entity confidence

Cost Sentinel

Variable	Default	Description
RAG_COST_TRACKING_ENABLED	True	Enable cost tracking
RAG_COST_ALERT_THRESHOLD_USD	100.0	Alert threshold

Risk Mitigations

Risk	Mitigation
Table structure loss	VLM fallback for complex tables
OpenSearch memory	Tuned ef_construction and m parameters
Orphaned chunks	Contextual enrichment (title+summary in every chunk)
Indexing bottlenecks	Batch processing, _bulk API, async
Neo4j connection failure	Graceful degradation — pipeline continues without graph
PII false positives	Tunable confidence threshold; human-in-the-loop
LLM API failure	Circuit breaker pattern with exponential backoff
Embedding model unavailable	Graceful fallback to zero vectors

Testing

# Run all tests
python -m pytest tests/ -v

# Run specific test suites
python -m pytest tests/test_knowledge.py -v
python -m pytest tests/test_compliance.py -v
python -m pytest tests/test_observability.py -v

# With coverage
python -m pytest tests/ --cov=src --cov-report=html

Contributing

Contributions are welcome! Please open an issue or pull request on GitHub.

License

MIT — see LICENSE for details.