📊 VectorDB From Scratch

Learning project: Building my own vector database to understand how they work. Implemented from scratch to trace every step.

---

Why I Built This

To understand how vector databases like Pinecone, Weaviate, and Chroma work by building one myself.

Goal: Learn by doing, not just using.

---

What I Learned

✅ Vector embeddings and similarity search
✅ TF-IDF vectorization from scratch
✅ Word embeddings (Word2Vec concepts)
✅ PCA for dimensionality reduction
✅ Visualizing high-dimensional data in 2D

---

What It Does

• Converts text to vector embeddings
• Stores vectors in memory
• Searches similar vectors using cosine similarity
• Visualizes embeddings with PCA

---

Quick Start

python wordEmbedding.py
# or
python tfidf.py

---

Files

• wordEmbedding.py - Word embedding implementation
• tfidf.py - TF-IDF vectorization
• Both include PCA visualization

---

How It Works

# 1. Convert text to vectors
vectors = create_embeddings(documents)

# 2. Store in "database" (Python dict)
db = VectorDatabase()
db.add(vectors)

# 3. Search similar vectors
results = db.search(query_vector, top_k=5)

# 4. Visualize with PCA
visualize_2d(vectors)

---

What Vector DBs Are Used For

• Semantic search (search by meaning, not keywords)
• Recommendation systems
• RAG (Retrieval Augmented Generation) for LLMs
• Image similarity search
• Duplicate detection

---

Key Concepts I Traced

Embeddings: Text → Numbers (vectors)
Similarity: Cosine distance between vectors
Indexing: Fast lookup structures
PCA: High dimensions → 2D for visualization

---

Tech Stack

• Python
• NumPy (vector operations)
• scikit-learn (PCA)
• Matplotlib (visualization)

---

Next Steps

• Add FAISS indexing for speed
• Implement HNSW algorithm
• Add persistence (save/load)
• Build REST API
• Compare with real vector DBs

---

Author

Goutham N
GitHub: @GOUTHAM-2002

---

⭐ Star if you're also learning by building!

---

Note: This is a learning project. For production use, check out Pinecone, Weaviate, or Chroma.