NepalElection.chat
AI Election Copilot — Built & Deployed in 24 Hours
About the Project
NepalElection.chat is a live AI election intelligence platform built and deployed in 24 hours during Nepal's 2082 federal election. It combines a multi-agent data pipeline scraping real-time results from Nepal's top news sources with a bilingual AI chat interface — giving any user instant answers about 165 constituencies, 3,406 candidates, and 77 districts in English or Nepali. Within 24 hours of launch: 13K Google Search impressions, 724 organic clicks, and users from 5 countries — all from zero.
The Challenge
Nepal's election results were scattered across multiple Nepali-language news sources with conflicting data, no unified API, and no English interface. Diaspora communities in the US, UK, and Australia had no reliable, real-time way to follow results. The window to be useful was narrow — election results only matter for hours.
The Solution
Built a three-layer agentic system in 24 hours. Layer 1: Data ingestion agents continuously scraping election.onlinekhabar.com (4.5K API calls) and election.ekantipur.com, reconciling discrepancies between sources and caching results in Upstash Redis (4.8K calls) for sub-second reads. Layer 2: A cron-driven refresh pipeline running on Vercel edge functions, keeping all 165 constituency results current without manual intervention. Layer 3: A bilingual AI chat agent (English + Nepali) that lets users ask natural language questions — "who won in Kathmandu?", "how many seats does Congress have?" — with cited sources and timestamps. An interactive SVG map of Nepal's 77 districts color-coded by winning party completed the experience.
Impact & Results
- 13K Google Search impressions and 724 organic clicks within 24 hours of launch — from zero prior search presence.
- Reached users across 5 countries organically: Nepal (56%), USA (13%), India (5%), UK (4%), Australia (3%) — the exact diaspora audience the tool was built for.
- Indexed and ranked for high-intent queries including "nepal election map", "nepal election live count", and "sudurpaschim election result 2082" with avg position 6.7.
- Zero-error data pipeline: 4.8K Redis cache hits and 4.5K live scrape calls processed across /results/[slug] (155 invocations), /api/hot-seats, and map routes with 0% error rate on all production routes.
- Demonstrated the full Forward Deployed AI loop: identify the moment → architect the agentic system → ship to production → real users in hours.