Spaces:
Running
Running
| description = """ | |
| ## 🕉️ **Project Title: RamayanaGPT & GitaGPT – RAG-based Chatbots for Ancient Indian Epics** | |
| --- | |
| ### 🔍 **Project Overview** | |
| **RamayanaGPT** and **GitaGPT** are knowledge-based conversational AI tools designed to answer questions from the *Valmiki Ramayana* and the *Bhagavad Gita*, respectively. These chatbots use **Retrieval-Augmented Generation (RAG)** architecture to generate accurate, scripture-based responses. They combine powerful **vector search capabilities** with **large language models (LLMs)** to deliver spiritually insightful, context-rich conversations. | |
| These tools leverage: | |
| * **MongoDB Atlas Vector Search** for embedding-based document retrieval | |
| * **Hugging Face** embeddings (`intfloat/multilingual-e5-base`) | |
| * **Groq LLaMA 3.1 8B** via API | |
| * **LlamaIndex** for orchestration | |
| * **Gradio** for user interface | |
| --- | |
| ### 🏗️ **Key Components** | |
| #### 1. **Vector Store: MongoDB Atlas** | |
| * Two collections are created in the `RAG` database: | |
| * `ramayana` for **Valmiki Ramayana** | |
| * `bhagavad_gita` for **Bhagavad Gita** | |
| * Each collection contains vector indexes: | |
| * `ramayana_vector_index` | |
| * `gita_vector_index` | |
| * Each document includes: | |
| * For Ramayana: `kanda`, `sarga`, `shloka`, `shloka_text`, and `explanation` | |
| * For Gita: `Title`, `Chapter`, `Verse`, and `explanation` | |
| #### 2. **Vector Embedding: Hugging Face** | |
| * Model: `intfloat/multilingual-e5-base` | |
| * Used to convert `shloka_text + explanation` or `verse + explanation` into vector representations | |
| * These embeddings are indexed into MongoDB for semantic similarity search | |
| #### 3. **Language Model: Groq API** | |
| * LLM used: `llama-3.1-8b-instant` via **Groq API** | |
| * Users input their Groq API key securely | |
| * LLM is instantiated per query using `llama_index.llms.groq.Groq` | |
| #### 4. **Prompt Engineering** | |
| * Custom **PromptTemplates** guide the response structure for each chatbot | |
| * **RamayanaGPT Prompt**: | |
| * Introduction to the query | |
| * Related shlokas with explanations | |
| * Summary/overview | |
| * **GitaGPT Prompt**: | |
| * Context or spiritual background | |
| * Relevant verse(s) with meaning | |
| * Reflective conclusion | |
| #### 5. **Index Initialization** | |
| * Vector indexes are loaded **once** at application startup: | |
| ```python | |
| ramayana_index = get_vector_index("RAG", "ramayana", "ramayana_vector_index") | |
| gita_index = get_vector_index("RAG", "bhagavad_gita", "gita_vector_index") | |
| ``` | |
| * Shared across all user queries for speed and efficiency | |
| #### 6. **User Interface: Gradio** | |
| * Built with `gr.Blocks` using the `Soft` theme and `Roboto Mono` font | |
| * Two tabs: | |
| * 🕉️ **RamayanaGPT** | |
| * 🕉️ **GitaGPT** | |
| * Users enter their Groq API key once; it's stored in `gr.State` | |
| * Upon authentication: | |
| * API key input and help accordion are hidden | |
| * Full chat interface is revealed (`gr.ChatInterface`) | |
| --- | |
| ### ⚙️ **Technical Stack** | |
| | Component | Technology | | |
| | --------------- | ------------------------------------- | | |
| | Backend LLM | Groq (LLaMA 3.1 8B via API) | | |
| | Embedding Model | Hugging Face (`multilingual-e5-base`) | | |
| | Vector Store | MongoDB Atlas Vector Search | | |
| | Vector Engine | LlamaIndex VectorStoreIndex | | |
| | Prompt Engine | LlamaIndex PromptTemplate | | |
| | Query Engine | LlamaIndex Query Engine | | |
| | UI Framework | Gradio (Blocks + ChatInterface) | | |
| | Deployment | Python app using `app.py` | | |
| --- | |
| ### ✅ **Features Implemented** | |
| * [x] Vector search using MongoDB Atlas | |
| * `ramayana_vector_index` for Valmiki Ramayana | |
| * `gita_vector_index` for Bhagavad Gita | |
| * [x] Hugging Face embedding (`e5-base`) integration | |
| * [x] API key input and session handling with `gr.State` | |
| * [x] LLM integration via Groq API | |
| * [x] Prompt templates customized for each scripture | |
| * [x] Tabbed interface for seamless switching between RamayanaGPT and GitaGPT | |
| * [x] Clean UX with collapsible Groq API key instructions | |
| * [x] Logging of each query with timestamp (for debugging/monitoring) | |
| """ | |
| groq_api_key = """ | |
| ### 🔑 How to Get a Groq API Key | |
| 1. **Go to** [https://console.groq.com/keys](https://console.groq.com/keys) | |
| 2. **Log in or Sign Up** for a Groq account. | |
| 3. Click **"API Keys"** from the dashboard. | |
| 4. Click **"Create Key"**, name it, and generate. | |
| 5. **Copy the API key** and store it securely. | |
| 6. **Paste** the key into the RamayanaGPT app to start chatting. | |
| --- | |
| ⚠️ **Don't share** your API key. Revoke and regenerate if needed. | |
| """ | |
| RamayanaGPT=''' | |
| ## 🕉️ **RamayanaGPT – Overview and Dataset Summary** | |
| ### 📖 **Introduction** | |
| **RamayanaGPT** is a RAG-based chatbot that draws upon the **Valmiki Ramayana**, the original Sanskrit epic, to answer user queries with reference to shlokas and their commentaries. It aims to offer precise, contextual, and respectful responses using advanced retrieval and generation technologies. | |
| ### 🗂️ **Dataset Structure** | |
| The uploaded Ramayana dataset includes the following columns: | |
| | Column | Description | | |
| | ------------- | ------------------------------------------------------------------------------ | | |
| | `kanda` | One of the 7 books (kandas) of the Ramayana (e.g., Bala Kanda, Ayodhya Kanda). | | |
| | `sarga` | The chapter number within each kanda. | | |
| | `shloka` | The shloka (verse) number within the sarga. | | |
| | `shloka_text` | Original Sanskrit verse. | | |
| | `explanation` | English explanation or interpretation of the shloka. | | |
| ### 🔍 **Example** | |
| ```text | |
| kanda: Bala Kanda | |
| sarga: 1 | |
| shloka: 1 | |
| shloka_text: तपस्स्वाध्यायनिरतं तपस्वी वाग्विदां वरम् । | |
| explanation: Ascetic Valmiki enquired of Narada, preeminent among sages, who was engaged in penance and study of the Vedas. | |
| ``` | |
| ### 💡 **Insights** | |
| * The data is well-structured with nearly **1,400+** records. | |
| * Each record reflects a deep philosophical or narrative moment from the epic. | |
| * Metadata (`kanda`, `sarga`, `shloka`) allows precise retrieval and organization. | |
| * Used for vector indexing and semantic retrieval. | |
| ''' | |
| GitaGPT=''' | |
| ## 🕉️ **GitaGPT – Overview and Dataset Summary** | |
| ### 📖 **Introduction** | |
| **GitaGPT** is a chatbot built to answer spiritual and philosophical questions using the **Bhagavad Gita** as its primary source. It references verses (slokas) directly from the Gita, delivering insights supported by both Sanskrit, Hindi, and English explanations. | |
| ### 🗂️ **Dataset Structure** | |
| The uploaded Gita dataset contains the following fields: | |
| | Column | Description | | |
| | --------------------- | --------------------------------------------------- | | |
| | `S.No.` | Serial number of the verse. | | |
| | `Title` | Title of the chapter (e.g., Arjuna's Vishada Yoga). | | |
| | `Chapter` | Gita chapter number (e.g., Chapter 1). | | |
| | `Verse` | Verse ID (e.g., Verse 1.1). | | |
| | `Sanskrit Anuvad` | Original verse in Devanagari Sanskrit. | | |
| | `Hindi Anuvad` | Hindi translation/interpretation. | | |
| | `Enlgish Translation` | English translation/interpretation. | | |
| ### 🔍 **Example** | |
| ```text | |
| Chapter: Chapter 1 | |
| Verse: Verse 1.1 | |
| Sanskrit: धृतराष्ट्र उवाच । धर्मक्षेत्रे कुरुक्षेत्रे समवेता युयुत्सवः... | |
| Hindi: धृतराष्ट्र बोले- हे संजय! धर्मभूमि कुरुक्षेत्र में एकत्र हुए युद्ध की इच्छा रखने वाले... | |
| English: Dhrtarashtra asked of Sanjaya: O SANJAYA, what did my sons and the sons of Pandu do? | |
| ``` | |
| ### 💡 **Insights** | |
| * The dataset contains **700+ verses** from all 18 chapters. | |
| * Multilingual representation (Sanskrit, Hindi, English) enhances usability for diverse users. | |
| * The verse structure (`Chapter`, `Verse`) aids in precise referencing and response generation. | |
| * Perfectly suited for semantic search via vector embeddings. | |
| ''' | |
| footer = """ | |
| <div style="background-color: #1d2938; color: white; padding: 10px; width: 100%; bottom: 0; left: 0; display: flex; justify-content: space-between; align-items: center; padding: .2rem 35px; box-sizing: border-box; font-size: 16px;"> | |
| <div style="text-align: left;"> | |
| <p style="margin: 0;">© 2025 </p> | |
| </div> | |
| <div style="text-align: center; flex-grow: 1;"> | |
| <p style="margin: 0;"> This website is made with ❤ by SARATH CHANDRA</p> | |
| </div> | |
| <div class="social-links" style="display: flex; gap: 20px; justify-content: flex-end; align-items: center;"> | |
| <a href="https://github.com/21bq1a4210" target="_blank" style="text-align: center;"> | |
| <img src="data:image/png;base64,{}" alt="GitHub" width="40" height="40" style="display: block; margin: 0 auto;"> | |
| <span style="font-size: 14px;">GitHub</span> | |
| </a> | |
| <a href="https://www.linkedin.com/in/sarath-chandra-bandreddi-07393b1aa/" target="_blank" style="text-align: center;"> | |
| <img src="data:image/png;base64,{}" alt="LinkedIn" width="40" height="40" style="display: block; margin: 0 auto;"> | |
| <span style="font-size: 14px;">LinkedIn</span> | |
| </a> | |
| <a href="https://21bq1a4210.github.io/MyPortfolio-/" target="_blank" style="text-align: center;"> | |
| <img src="data:image/png;base64,{}" alt="Portfolio" width="40" height="40" style="display: block; margin-right: 40px;"> | |
| <span style="font-size: 14px;">Portfolio</span> | |
| </a> | |
| </div> | |
| </div> | |
| """ |