AI Powered Standards Recommendation Engine

Maanak

AI-Powered Recommendation & Search Engine for Indian Standards (BIS)

Maanak is an intelligent Retrieval-Augmented Generation (RAG) platform designed to answer queries regarding Bureau of Indian Standards (BIS). Instead of relying on a language model's pre-trained knowledge—which can lead to hallucinations or outdated information—Maanak retrieves verified standard-specific data prior to generating precise, structured answers grounded in actual document content.

---

User Interface

---

Overview & System Architecture

Maanak operates across three primary decoupled pipelines: Data Augmentation, Vector Retrieval, and Contextual Generation.

+-----------------------------------------------------------------------------------+
|                            1. DATA AUGMENTATION PIPELINE                          |
|  [ BIS Official Data ] -> [ Data Cleaning ] -> [ Structuring ] -> [ Embeddings ]  |
+-----------------------------------------------------------------------------------+
                                          |
                                          v
+-----------------------------------------------------------------------------------+
|                              2. RETRIEVAL PIPELINE                                |
|  [ User Query ] -> [ Query Embedding ] -> [ Qdrant DB Search ] -> [ Top-5 Chunks] |
+-----------------------------------------------------------------------------------+
                                          |
                                          v
+-----------------------------------------------------------------------------------+
|                             3. GENERATION PIPELINE                                |
|  [ Top-5 Chunks + Query ] -> [ Prompt ] -> [ Grok LLM ] -> [ Structured Answer ]  |
+-----------------------------------------------------------------------------------+

---

Technical Pipeline Breakdown

1. Data Augmentation Pipeline

• Data Collection: Collects raw Indian Standards data directly from official sources.
• Data Cleaning & Structuring: Irrelevant noise is removed, and meaningful content is organized into a structured JSON format.
• Embedding Generation: Structured entries are processed through a sentence-transformers model to generate high-dimensional numerical vector representations that capture semantic meaning.

2. Retrieval Pipeline

• Vector Storage: Embeddings and metadata are stored in Qdrant Vector Database, supporting combined semantic similarity and keyword-based filtering.
• Query Embedding: Incoming user queries are embedded using the exact same sentence-transformer model to map into the shared vector space.
• Top-K Retrieval: Executes hybrid search in Qdrant to pull the top 5 (k=5) most relevant context entries for the given query.

3. Generation Pipeline

• Prompt Engineering: Merges the retrieved top-5 context entries with the original query into a structured system prompt.
• LLM Inference: Sends the context-rich prompt to the Grok (xAI) language model.
• Response Formatting: Post-processes the raw model output into a standardized layout for presentation.

---

Key Benefits

• High Accuracy: Reduces hallucinations by grounding answer generation inside official BIS documentation.
• Scalable Knowledge Base: New standards can be ingested into Qdrant independently without retraining or fine-tuning models.
• Fast Retrieval: Qdrant vector indexing isolates only the relevant information entries instead of processing raw document trees.
• Decoupled Architecture: Modular design allows swapping generation models or vector databases without altering frontend or pipeline logic.

---

Directory Structure

Maanak/
├── .github/
│   └── workflows/          # CI/CD automation and test pipelines
├── backend/                # FastAPI (Python) Service
│   ├── app/
│   │   ├── api/            # Route endpoints and request validation
│   │   ├── core/           # Configuration and environment management
│   │   ├── config/         # Configuration Files
│   │   ├── schema/         # Hold the Data transfer Object files
│   │   └── service/        # Transformer embeddings and LLM integrations
│   ├── tests/              # Unit and integration test suites
│   └── requirements.txt    # Python dependencies
├── frontend/               # Next.js (TypeScript) Client
│   ├── public/             # Static assets, branding, and images
│   ├── src/
│   │   ├── components/     # UI elements (search, standards viewer, footer)
│   │   ├── pages/          # Application routes and dynamic views
│   │   ├── services/       # API integration client
│   │   └── styles/         # Global styles and layout themes
│   ├── package.json        # Frontend scripts and dependencies
│   └── tsconfig.json       # TypeScript configuration
├── extension/              # Chrome browser extension module
├── CONTRIBUTIONS.md        # Contribution guidelines
└── README.md               # Project documentation

---

Architecture

Getting Started

Prerequisites

• Python 3.9 or higher
• Node.js v18 or higher
• Running instance of Qdrant Vector Database (Local or Cloud)
• API Key for Groq

---

Backend Setup

1. Change directory into backend:

cd backend

2. Create and activate a virtual environment:

uv init
source venv/bin/activate

3. Install required Python packages:

uv add -r requirements.txt

4. Configure environment variables (create a .env file in backend/):

QDRANT_SERVER_URL=http://localhost:6333
QDRANT_API_KEY=qdrant_api_key_if_using_cloud_version
QDRANT_COLLECTION_NAME=bis_standards
GROQ_API_KEY=groq_api_key

5. Start the FastAPI backend server:

uv run python -m app.main

---

Frontend Setup

1. Open a new terminal and navigate to frontend:

cd frontend

2. Install client dependencies:

npm install

3. Configure local environment variables (create .env.local in frontend/):

NEXT_PUBLIC_API_URL=http://localhost:8080/api

4. Run the development server:

npm run dev

Access points:

• Web Client: http://localhost:3000
• API Documentation: http://localhost:8080/docs

---

Thank You

Thank you for exploring Maanak. Contributions, feedback, and pull requests are welcomed to help advance automated access to Indian Standards.