🧭 DocuVector AI

In-Browser High-Dimensional Vector Embeddings & Semantic Cosine Retrieval Engine

---

⚡ Problem Statement

Keyword-matching search algorithms (like browser Ctrl+F or standard string grep) evaluate strictly identical characters. When users search for conceptual meaning (e.g. asking "How does memory compression improve model latency?" across a document using phrases like "quantization minimizes memory bandwidth"), traditional search returns zero hits.

DocuVector AI is a client-side semantic document intelligence engine created by Krishna Gupta. It constructs a high-dimensional vector space from arbitrary text, applies sublinear TF-IDF term weighting, and executes instantaneous cosine similarity scoring with interactive 2D vector space projection running entirely in the client's browser with zero backend servers and zero API costs.

---

📐 Mathematical Formulation

The semantic proximity between the user query vector $\vec{q}$ and any document chunk vector $\vec{d}_i$ is computed via normalized dot-product cosine distance:

$$\text{Cosine Similarity}(\vec{q}, \vec{d}_i) = \frac{\vec{q} \cdot \vec{d}_i}{|\vec{q}|_2 |\vec{d}_i|_2} = \frac{\sum_{j=1}^{N} q_j \cdot d_{i,j}}{\sqrt{\sum_{j=1}^{N} q_j^2} \sqrt{\sum_{j=1}^{N} d_{i,j}^2}}$$

Where:

• $N$ represents the active corpus vocabulary dimension.
• Each weight incorporates Inverse Document Frequency: $\text{IDF}(t) = \ln\left(\frac{N_{docs} + 1}{DF(t) + 0.5}\right) + 1$.
• Vectors are $L_2$-normalized such that $|\vec{v}|_2 = 1.0$.

---

🛠️ Architecture & Pipeline

flowchart LR
    A["📄 Raw Unstructured Document / Notes"] --> B["✂️ Recursive Paragraph Chunker"]
    B --> C["🔤 Tokenizer & Stopword Filtration"]
    C --> D["📐 High-Dimensional TF-IDF Matrix"]
    D --> E["⚡ L2 Normalized Chunk Vectors"]
    F["🔍 Natural Language Concept Query"] --> G["🧮 Query Vector Transformation"]
    E --> H["🎯 Cosine Similarity Dot-Product Engine"]
    G --> H
    H --> I["📊 Top-K Ranked Semantic Results"]
    H --> J["🌐 2D Polar Vector Space Projection Canvas"]

Loading

---

✨ Key Features

• Sub-Millisecond In-Browser Vectorization: Generates high-dimensional vector spaces and computes cosine scores across dozens of chunks in < 2 ms.
• Interactive 2D Vector Radar: Maps multi-dimensional chunk vectors and the query vector onto a 2D polar projection canvas, rendering real-time vector connection lines.
• Dynamic Concept Highlighting: Automatically detects and highlights relevant keyword tokens inside retrieved passages.
• Pre-Loaded Multi-Domain Corpora: 1-click test datasets for AI & Machine Learning Research, Clinical Medical Trials, and Distributed Cloud Architecture.

---

🚀 Running Locally

1. Clone this repository:

   git clone https://github.com/KrishnaGupta-coder/DocuVector-AI.git
   cd DocuVector-AI

2. Open index.html in any browser, or run via Python:

   python -m http.server 8000

3. Ingest documents and explore semantic similarity!

---

_{Developed by Krishna Gupta • Specialized in Vector Similarity & Machine Learning Systems • Jaipur, India}