Technical Overview

How Ekantik Vartalap Archive Works

यह प्रोजेक्ट तकनीकी रूप से कैसे काम करता है?

Learn how this project uses semantic search to help you find relevant questions and moments from Maharaj Ji's original videos.

It does not generate spiritual answers using an AI chatbot. Instead, it searches existing questions and connects them with the corresponding YouTube videos and timestamps.

How Search Works

सर्च कैसे काम करता है?

1

You ask a question

You enter a question in Hindi, English, or Hinglish.

"How can I deal with anxiety?"
2

Your question is converted into an embedding

A Python background worker uses the embedding model to convert your question into a numerical representation of its meaning. (Currently it is being run on my laptop.)

3

Similar questions are found

The embedding is compared with embeddings of existing questions stored in the database.

4

Relevant video moments are matched

The most relevant questions are connected to their corresponding videos and timestamps.

5

Results are shown to you

The relevant YouTube video moments are returned with their timestamps so you can directly watch the original content.

Technology Stack

तकनीकी संरचना

Next.js

Full-Stack Framework

Next.js is used as the full-stack framework for the project. It handles both the frontend website and the backend APIs.

PostgreSQL + pg_vector

Database & Semantic Search

PostgreSQL is used as the database, with the pg_vector extension enabled to store and search embeddings.

Python Background Worker

Embedding & Matching

A Python background worker creates embeddings for user queries, finds similar questions, and stores the search results in the database.

Ingestion Pipeline

Preparing the Search Data

An ingestion pipeline processes the existing questions, creates their embeddings, and stores them in the database for future searches.

How Questions Are Added

प्रश्नों को कैसे तैयार किया जाता है?

Before users search, the existing questions need to be prepared for semantic search. This is done using an ingestion pipeline.

1

Collect questions from the available Ekantik Vartalap content.

2

Process and prepare the questions.

3

Create embeddings for the questions.

4

Store the embeddings in PostgreSQL.

5

Use these embeddings for future searches.

Python Background Worker

बैकग्राउंड प्रोसेसिंग

The heavier embedding and matching work is handled by a Python background worker. This keeps the web application separate from the processing work.

✓

Create embeddings for user queries

✓

Find similar existing questions

✓

Match questions with relevant content

✓

Store search results in PostgreSQL

Embedding Model

एम्बेडिंग मॉडल

Model

bhasha-embed-v0

The project currently uses bhasha-embed-v0 to create embeddings for questions and search queries.

A special thanks to the creator for making this embedding model available.

Simple Architecture

प्रोजेक्ट की संरचना

User
Next.js Website
Next.js Backend
Search Request Queue
Python Background Worker
Bhasha-Embed Model
PostgreSQL + pg_vector
Similar Questions
Video + Timestamp Results

Why It Works This Way

इसका उद्देश्य

The goal of this project is simple: help people find original teachings that are relevant to their questions.

The website does not try to create new spiritual answers. Instead, it uses semantic search to help people discover relevant moments from the original videos.

राधे राधे 🙏