I.Q.R.S
Intelligent Query Retrieval System
RAG (Retrieval-Augmented Generation) | Document Intelligence
Overview
I.Q.R.S. is an end-to-end Retrieval-Augmented Generation platform that converts unstructured PDF documents into a searchable intelligence layer. Users ask natural-language questions and receive fact-grounded answers with traceable evidence – not generic chatbot responses.
Key Features
PDF Ingestion Pipeline
Page-wise text extraction with multi-PDF support.
Smart Chunking
Sentence-level chunking with configurable overlap for semantic continuity.
Index Persistence
FAISS index and chunk store cached to disk. Avoids reprocessing.
Vector Similarity Search
Top-K relevant chunks retrieved via dense embedding similarity.
Grounded Answer Generation
LLM restricted to retrieved evidence. Hallucination suppressed.
Evidence Traceability
Page + line markers, chunk IDs, similarity scores with every answer.
Technology Stack
Algorithms & Methods
Document Chunking
Sentence-based segmentation with word-count bounds and configurable overlap.
E5-small-v2 Embeddings
intfloat/e5-small-v2 → dense vector embedding for chunks and queries.
FAISS Cosine Similarity
Query vs chunk embedding similarity → Top-K retrieval.
Grounded Generation (Groq)
LLM with strict instruction to answer only from context.
System Architecture
PDF Processor
PDF parsing, chunk generation, embedding, FAISS index creation
QA Engine
Load index, embed query, retrieve Top-K, generate grounded response
API Layer
Query endpoint routing and orchestration
Web Interface
User-facing query UI
Results & Outcomes
- Full-stack RAG pipeline from PDF ingestion to grounded QA
- FAISS-backed retrieval with sub-second latency
- Auditable answers with page/line evidence markers
- Production-deployable FastAPI architecture