Skip to main content
The Standard RAG (Retrieval-Augmented Generation) System in Avigrah provides a structured way for AI agents to access and reason over large volumes of data. It ensures that every AI response is grounded in retrieved, relevant, and contextually accurate information, rather than relying solely on model memory.

​
What is RAG?

Retrieval-Augmented Generation (RAG) is a system where:
  • Data is stored in a searchable format
  • Relevant information is retrieved at runtime
  • AI generates responses based on that retrieved context
This approach ensures that outputs are:
  • Accurate
  • Context-aware
  • Aligned with your actual data

​
How the Standard RAG System Works

The RAG system in Avigrah follows a structured pipeline: Data → Embedding → Vector Storage → Retrieval → AI Response

​
1. Data Processing

Data from pipelines is:
  • Cleaned and structured
  • Converted into smaller chunks
  • Prepared for embedding
This ensures that information is searchable and retrievable.

​
2. Embedding Model

Avigrah uses: all-MiniLM-L6-v2 (default) This model converts text into vector representations. These vectors:
  • Capture semantic meaning
  • Enable similarity-based search
  • Allow AI to retrieve relevant information even when queries are not exact matches

​
3. Vector Database

All embeddings are stored in: Qdrant Vector Database Qdrant enables:
  • High-performance vector search
  • Scalable storage of embeddings
  • Fast retrieval during AI execution
Avigrah uses: Cosine Similarity This method compares vectors based on their semantic closeness.
  • Higher similarity → More relevant context
  • Lower similarity → Less relevant
This ensures that AI retrieves the most meaningful data for each query.

​
5. Context Retrieval

When an AI agent executes:
  • The query is converted into a vector
  • Relevant data points are retrieved from Qdrant
  • Top matching results are selected
This retrieved data becomes the context for the AI response.

​
6. AI Response Generation

The AI model receives:
  • The user query
  • Retrieved context from the vector database
It then generates a response that is:
  • Grounded in actual data
  • Contextually accurate
  • Consistent with the underlying information

​
Why This System Matters

The Standard RAG System solves one of the biggest limitations of AI systems:

​
Hallucination

Without retrieval:
  • AI guesses based on training data
With RAG:
  • AI responds based on actual retrieved data

​
Key Characteristics

Finds meaning, not just keywords

​
Scalable Architecture

Handles large datasets efficiently

​
Low Latency Retrieval

Optimized for fast response times

​
Consistent Outputs

Same query → same context → reliable answers

​
Where RAG is Used

The Standard RAG System is used for:
  • Document understanding
  • Knowledge base systems
  • Historical data analysis
  • Context-aware AI responses
  • Retrieval-driven reasoning

​
Integration with Pipelines

RAG is tightly integrated with data pipelines.
  • Pipelines prepare and structure data
  • Data is vectorized and stored in Qdrant
  • AI agents query this data during execution
This ensures a continuous flow from data → intelligence.

​
Default Configuration

Avigrah’s Standard RAG system is optimized out of the box:
  • Embedding Model: all-MiniLM-L6-v2
  • Vector Database: Qdrant
  • Similarity Metric: Cosine Similarity
These defaults provide a balance of:
  • Performance
  • Accuracy
  • Scalability

​
Positioning

The Standard RAG System is the core retrieval layer of Avigrah. It enables AI agents to move beyond static knowledge and operate on real, dynamic, and structured data. This is what makes Avigrah’s AI systems:
  • Reliable
  • Context-aware
  • Production-ready