• 

  Gram setu Ai

Inspiration

Family and field staff in small towns juggle agriculture, energy bills, and info gaps across languages. We wanted one simple assistant that works offline-first and cites sources so people can trust it. Centre for Networked Intelligence

What it does

Agri agent: season-aware crop playbooks and basic advisory.

Energy agent: practical tips and mini load-management playbooks.

Sahay agent: schemes/helplines/logistics checklists. Agents coordinate via an Orchestrator and always show sources.

How we built it

Modular agents (Agri, Energy, Sahay) coordinated by an orchestrator. Granite models for reasoning; RAG over vetted Indian content; IBM Agent Development Kit (ADK) for packaging and workflows. CLI demo today; chat interface next.

Challenges we ran into

:Robust multilingual RAG in low-connectivity settings.

:Balancing transparency with short, actionable replies.

:Clean agent hand-offs without hard-wiring to one model family.

Accomplishments we’re proud of

Clean, extensible agent skeleton with clear contracts.

Lightweight CLI that runs on low-resource laptops.

Print-ready proposal + submission assets prepared.

What we learned

Agent UX > raw model scores.

Offline-first + citations drive trust.

Guardrails = policy + data, not only prompts.

What’s next

Wire in ADK routing and Granite calls end-to-end.

Add curated corpora and evaluations.

Ship WhatsApp bot; test with 5–10 field users.

GramSetu - AI for Rural India

Built With

• faiss/chroma-(rag)
• fastapi/cli
• granite-3.x-(open)
• ibm-watsonx-orchestrate-adk
• langchain
• numpy/pandas
• pydantic
• python
• transformers