Logo Lanfrica

lonestarx1/powerpulse-africa

Domain:

environment and energydigital infrastructure

Record type:

softwareproject
Creator:
lon
Host:
The power just went out. Now what? # PowerPulse **Ask about a power outage in your own language and get a straight answer: is it planned, how long, and what to do right now.** Built at the Frontiers GenAI Hackathon, ALX Kigali, 28 July 2026 — Track 04, Energy & Climate Infrastructure. --- ## The problem Across much of Africa, when the power goes out people are left asking the same three questions: *Is this planned? How long will it last? What do I do now?* Utilities often publish the answers — but in English, in paginated HTML tables that are painful to use on a phone, and with no guidance attached. Rwanda Energy Group's outage listing runs to **126 pages**. Nobody sitting in the dark is going to page through it. The third question is the one nobody answers at all. A two-hour planned maintenance window and an open-ended fault call for completely different decisions, and those decisions differ again depending on whether you are running a household, protecting cold stock in a shop, or trying to stay online for work. ## What it does You type where you are, in English, Kinyarwanda, or a mix. PowerPulse: 1. **Matches your message to a real place** in the utility's own records — including misspellings and code-switched phrasing. 2. **Tells you what the utility actually published**: planned or not, the window, the stated cause, the feeder. 3. **Grounds it in history** — how often this sector loses power, for how long, and on how many past records that is based. 4. **Gives advice conditioned on the duration and on who you are.** Same place, same moment, different profile → materially different guidance. 5. **Says "I don't know" when the data cannot support an answer.** No invented restoration times, ever. Rwanda is the MVP. The architecture is utility-agnostic: expanding to another country means adding that utility's data source and language, not rebuilding the product. ## Why generative AI is load-bearing Not decoration — three places where nothing else works: - **Messy, code-switched input …