ApexRetriever

A lightweight hybrid retrieval system designed for fast semantic search and RAG pipelines.

Built for:

• semantic search
• lightweight RAG
• AI assistants
• retrieval systems
• local document QA

---

Architecture

Stage ① — BM25 Sparse Retrieval

Keyword-based retrieval for fast lexical matching.

Stage ② — Dense Semantic Search

Powered by:

• BAAI/bge-small-en-v1.5

Uses FAISS vector indexing.

Stage ③ — CrossEncoder Reranking

Final neural reranking using:

• cross-encoder/ms-marco-MiniLM-L-6-v2

---

Features

• Hybrid retrieval
• Fast indexing
• Dense semantic search
• Neural reranking
• Lightweight deployment
• GPU acceleration
• FAISS support
• Easy integration

---

Repository Structure

ApexRetriever/
│
├── bi_encoder/
├── reranker/
├── pipeline.py
└── README.md

---

Installation

pip install -U \
    sentence-transformers \
    transformers \
    faiss-cpu \
    rank-bm25 \
    torch

---

Quick Start

from pipeline import ApexRetriever

retriever = ApexRetriever(model_dir=".")

# Example documents

docs = [
    "Python was created by Guido van Rossum.",
    "Paris is the capital of France."
]

retriever.index_documents(docs)

results = retriever.retrieve(
    "Who created Python?",
    top_k=3
)

print(results)

---

Use Cases

• RAG systems
• Semantic search
• AI chatbots
• Knowledge retrieval
• Local search engines
• Memory systems

---

Performance

Recommended:

• CUDA GPU
• 8GB+ RAM
• Python 3.10+

---

Components

Component	Model
Dense Encoder	BAAI/bge-small-en-v1.5
Reranker	cross-encoder/ms-marco-MiniLM-L-6-v2
Vector Engine	FAISS
Sparse Search	BM25

---

License

Apache 2.0

---

QuantaSparkLabs