OAN Architecture

The Open Agriculture Network (OAN) architecture comprises a multilingual conversational AI and a Beckn-based open network. The AI enables voice/text user interaction in vernacular languages via language detection, ASR, moderation, Agentic LLM processing (vector search, API integration, glossary search), and TTS response. The Beckn protocol-powered network tackles agricultural fragmentation by providing digital infrastructure for information, transactions, and resource discovery. Together, they offer a comprehensive, integrated agriculture solution.

 

AI Module

This architecture illustrates a multilingual conversational AI system that allows a user to interact using text or voice queries in vernacular languages. The system processes these queries through language detection, moderation, retrieval, external APIs for external dynamic data and response delivery in a seamless pipeline.

The response again is constructed in voice and text in vernacular languages.

 

🔁Key Components and Flow

a) User Interaction

  • Input: The user can input his query in text or speak a query using their native language.

  • Voice Query: The spoken input is captured as a voice query.

b) Language Processing – Bhashini Platform

  • Language Detection + ASR:

    • The Bhashini module detects the spoken language and converts the voice input into a text query using Automatic Speech Recognition (ASR).

c) Agentic LLM (Large Language Model)

  • Moderation:

    • The query is passed through moderation to ensure safety and compliance.

  • Agentic LLM:

    • Vector Search: Searches relevant information from a vector database.

    • APIs: Fetches real-time data from external sources (e.g., weather, warehouse information, market price information).

    • Glossary Search: Looks up glossary or domain-specific terms.

    • Context Aggregation: Aggregates retrieved information for coherent response generation.

    • Intent: the user intent is captured from the query for processing

    • Main Agent Processing includes:

d) Data Ingestion and Management for Knowledge Bases

  • Moderated Response: Vector Database for Vector Search and Glossary Search for domain-specific terms as part of the Main Agent Processing.

e) External API Integration and Management Layer

  • Dynamic Data integration: The system uses "APIs (Weather, Warehouse, etc.)" to fetch real-time data.

  • External APIs with complex orchestrations are integrated, secured, authenticated, monitored for uptime.

f) Response Delivery

  • Moderated Response: The LLM’s output is moderated if needed.

  • TTS – Bhashini:

    • The AI response is converted back to the user’s language as voice output using Bhashini’s Text-to-Speech (TTS) service.

  • Streaming: The response is streamed to the user as audio output.

g) Feedback Loop for Continuous Improvement

  • Feedback loop: The current system collects user feedback, evaluating response quality, incorporating insights back into the system to continuously improve the LLM's performance, accuracy of responses, or the relevance of retrieved information.


Core Capabilities

  • ✅ Multilingual voice/text interface
  • ✅ Bhashini-powered ASR + TTS
  • ✅ Vector search for domain knowledge
  • ✅ Real-time external API data integration
  • ✅ Moderation pipeline

Advantages

  • 🌍 Vernacular support for all Indian languages
  • 🔒 Safe, moderated conversations
  • 📡 Live agricultural data (market, weather, warehouse)
  • 🔄 Continuous learning with feedback loop
  • 📱 Seamless voice-based experience

Network 

The agriculture ecosystem today consists of siloed, target specific solutions, which often fail to address systemic needs. Fragmented and non-interoperable solutions impede the seamless integration of stakeholders and resources. In order to synchronize the information flow, transaction flows and physical flows of goods and services between stakeholders, OAN envisions a transformative solution, adopting an infrastructure led approach : a foundational digital infrastructure that enables shared flow of value.

 

This digital infrastructure takes the form of an open network powered by Beckn protocol, which is a specification for resource discovery and transactions. OAN can be viewed as a grid of various interconnected groups of agri eco-system entities, stitched together through a unified scalable network infrastructure. These groups can enable independent policies for discovery, transaction, fulfillment and post fulfilment, through machine-readable templates. Simply put, this means when you want to buy something, the first step is finding it, this could be searching online or asking around (that's discovery). Once you’ve found what you need, you pay for it (that’s the transaction). Then, the item or service is delivered or provided to you (that’s fulfillment). After that, if you have any issues, need help, or want to give feedback, that’s what happens after the purchase (post-fulfillment).

 

The network can also leverage other building blocks such as universal identity systems (for individuals and assets - farms, machinery, crops, etc), machine readable data formats (going beyond static PDFs) and verifiable data packets. Such standards help in improving the trust in the network.

 

OAN can be described as synchronised setup of multiple layers:


An interoperable grid with all stakeholders across these various layers can coordinate, complement and unify existing solutions, while enabling innovations across the ecosystem.