Skip to main content

​
Overview

The RAG Recruitment Assistant uses HuggingFaceEmbeddings from LangChain to convert resume text into vector representations for semantic search and retrieval.

​
Installation

pip install langchain-huggingface

​
Basic Setup

from langchain_huggingface import HuggingFaceEmbeddings

# Initialize embeddings model
embeddings = HuggingFaceEmbeddings(
    model_name="sentence-transformers/all-MiniLM-L6-v2"
)

​
Configuration Parameters

​
model_name
string
default:"sentence-transformers/all-MiniLM-L6-v2"
The HuggingFace model identifier for generating embeddings.Common Models:
  • sentence-transformers/all-MiniLM-L6-v2 - Fast, lightweight (90MB)
  • sentence-transformers/all-mpnet-base-v2 - Higher quality, larger
  • sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 - Multilingual support
​
model_kwargs
dict
default:"{}"
Additional keyword arguments passed to the model.
embeddings = HuggingFaceEmbeddings(
    model_kwargs={'device': 'cuda'}  # Use GPU
)
​
encode_kwargs
dict
default:"{}"
Arguments for the encode method.
embeddings = HuggingFaceEmbeddings(
    encode_kwargs={'normalize_embeddings': True}
)
​
cache_folder
string
Directory to cache downloaded models.
embeddings = HuggingFaceEmbeddings(
    cache_folder="./model_cache"
)

​
Methods

​
embed_documents()

Generate embeddings for multiple documents.
​
texts
list[str]
required
List of text documents to embed.
​
embeddings
list[list[float]]
List of embedding vectors, one per input document.
documents = [
    "Estudiante de 9no ciclo con Python y React",
    "Data Analyst con experiencia en PowerBI"
]

vectors = embeddings.embed_documents(documents)
print(f"Generated {len(vectors)} embeddings")
print(f"Vector dimension: {len(vectors[0])}")
# Output: Generated 2 embeddings
# Output: Vector dimension: 384

​
embed_query()

Generate embedding for a single search query.
​
text
str
required
Query text to embed.
​
embedding
list[float]
Single embedding vector for the query.
query = "Busco desarrollador Python con FastAPI"
query_vector = embeddings.embed_query(query)
print(f"Query vector dimension: {len(query_vector)}")
# Output: Query vector dimension: 384

​
Real-World Usage

​
From the Talent Scout System

This example shows how embeddings are used in the actual recruitment assistant:
reference/notebook/Talent_Scout_3000x.ipynb
from langchain_community.document_loaders import PyPDFLoader
from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEmbeddings

# Initialize embeddings
embeddings = HuggingFaceEmbeddings()

# Load resume PDF
loader = PyPDFLoader("CV_Estudiante_4_Fernanda_Paredes.pdf")
docs = loader.load()

# Create vector store with embeddings
vectorstore = FAISS.from_documents(docs, embeddings)

​
Batch Processing Multiple Resumes

import glob
from langchain_community.document_loaders import PyPDFLoader

# Process multiple student resumes
archivos = glob.glob("cvs_estudiantes_final/*.pdf")

for pdf_path in archivos:
    loader = PyPDFLoader(pdf_path)
    pages = loader.load()
    
    # Generate embeddings for all pages
    texts = [p.page_content for p in pages]
    vectors = embeddings.embed_documents(texts)
    
    print(f"Processed {pdf_path}: {len(vectors)} vectors")

​
Performance Considerations

Model Download

On first run, the model (~90MB) is downloaded from HuggingFace Hub. Subsequent runs use the cached version.

GPU Acceleration

# Enable GPU for faster embedding generation
embeddings = HuggingFaceEmbeddings(
    model_kwargs={'device': 'cuda'}
)

Batch Size

For large documents, process in batches to avoid memory issues:
def batch_embed(texts, batch_size=32):
    all_embeddings = []
    for i in range(0, len(texts), batch_size):
        batch = texts[i:i + batch_size]
        embeddings_batch = embeddings.embed_documents(batch)
        all_embeddings.extend(embeddings_batch)
    return all_embeddings

​
Error Handling

import warnings
from langchain_huggingface import HuggingFaceEmbeddings

try:
    embeddings = HuggingFaceEmbeddings(
        model_name="sentence-transformers/all-MiniLM-L6-v2"
    )
    
    # Test embedding
    test_vector = embeddings.embed_query("Test query")
    print(f"✓ Embeddings loaded successfully ({len(test_vector)} dimensions)")
    
except Exception as e:
    print(f"Error loading embeddings: {e}")
    # Fallback to default model
    embeddings = HuggingFaceEmbeddings()

​
Vector Dimensions

ModelDimensionSizeSpeed
all-MiniLM-L6-v238490MBFast
all-mpnet-base-v2768420MBMedium
paraphrase-multilingual-MiniLM-L12-v2384420MBMedium

​
Next Steps

Vector Store

Learn how to store and search embeddings with FAISS

LLM Integration

Connect embeddings to the language model

Build docs developers (and LLMs) love

Get started for freeTalk to us