Gram Samachar

search bar for village

Inspiration

In a country where 500 million people live in rural areas, information is still a luxury. We realized that while India is digitizing rapidly, 67% of our 6.4 lakh villages are "digital deserts"—no local journalists, no websites, and no voice. We were inspired by the PESA areas (Scheduled Areas) where media coverage is nearly zero. We wanted to build a bridge that turns "Open Government Data" into "Local Village News" using the power of Gemini’s Grounding.

What it does

Gram Samachar is a "Cold-Start" News Engine. It doesn't need a database or prior training. A user simply enters a village name. In under 10 minutes, the pipeline fetches live data from e-Gram Swaraj (budgets), data.gov.in (market prices), and Satellite imagery (weather/crops). It generates a 5-point weekly news brief in the local dialect (Marathi, Hindi, etc.) at a Class 5 literacy level, ensuring everyone from a child to a grandparent can understand it.

How we built it

We built a sophisticated RAG (Retrieval-Augmented Generation) pipeline using: Gemini 3.1 Pro: Chosen for its massive context window and native Google Search Grounding. Python Backend: To automate the fetching of Panchayat PDF documents from government portals. Document Understanding: We used Gemini’s multimodal capabilities to "read" complex government budget tables and extract active construction projects. Prompt Engineering: We crafted a "System Role" that enforces a Class 5 reading level and specific local dialect output.

Challenges we ran into

The Data Gap: Most tiny villages have zero news mentions. We overcame this by shifting our focus from social news to governance data—treating budget approvals as the "headline."

Cold-Start Latency: Processing PDFs and live searches usually takes time. We optimized the pipeline to focus only on "Delta" (recent changes) to stay under the 10-minute constraint.

Language Nuance: Making the AI sound like a local neighbor instead of a robotic translator required deep iterative testing with few-shot prompting.

Accomplishments that we're proud of

Zero-Shot Accuracy: We successfully generated accurate news for a remote village in the Gadchiroli district without having any previous data on it.

Complex Data Simplification: Turning a 40-page technical Panchayat budget PDF into a single, simple sentence like "The village well repair has been approved" was a huge win.

Grounding Reliability: By using Gemini's grounding tool, we eliminated "hallucinations," ensuring every news point is backed by a government source.

What we learned

We learned that "Grounding" is the future of AI. You don't need to train a model on everything; you just need to teach it how to find everything. We also realized the importance of Inclusive Design—building for the "slow learner" or the low-literacy user actually makes the product better for everyone.

What's next for Gram Samachar

Voice-First Integration: We plan to integrate Gemini Live or a Text-to-Speech (TTS) engine so villagers can "listen" to the news while working in the fields.

Hyper-Local Weather Deltas: Using satellite APIs to alert farmers about specific crop diseases based on local humidity levels.

WhatsApp Bot: Deploying the pipeline as a WhatsApp bot so it can reach the 500M+ residents on the platform they use most