About FloodLine
When Chennai floods, phone lines go down and people ask for rescue on WhatsApp, X and Instagram. Volunteer groups read those messages by hand and pass them to boat teams. It works, but the same few things go wrong every time:
- One family gets posted ten times by different relatives, and two boats go to the same house.
- Nobody knows which requests are already handled, so old requests keep circulating after the family is safe.
- A request nobody picked up scrolls out of view, and nobody notices.
FloodLine is a shared list for that job. Volunteers paste the messages in. It pulls out the area, phone number, number of people, water level and who is vulnerable; merges repeats of the same household; puts the most urgent at the top with the reasons written out; and keeps each request on the list until someone marks the family as rescued.
It has been done before
In the 2018 Kerala floods, volunteers built keralarescue.in, an open-source site for rescue requests that the state adopted as its official one. FloodLine is a much smaller attempt at the same idea for Chennai, built for the way requests actually arrive here — as forwarded WhatsApp messages in Tamil, English and Tanglish.
How it decides what is urgent
Points, not a black box: pregnant or in labour +40, needs medical care +40, neck-deep water +35, baby +25, elderly +20, cannot walk +20, stuck on a terrace +20, plus time waiting with no one assigned and how many times it was reported. Every request shows its points. A volunteer can always overrule the order.
Where the AI is used, and where it isn't
Messages are first read by plain rules that work offline and never make things up. If a volunteer adds a Gemini key, Gemini re-reads each message to catch what rules miss ("near Phoenix mall" is Velachery; "old couple" is two people). Anything Gemini returns is checked against the original message: a phone number that isn't in the text, an area that isn't on the map or an address nobody wrote is thrown away, and the request says so.
What it can't do yet
- Boards sync between windows on one device. Different phones need a small server, which isn't built yet.
- It knows 28 flood-prone Chennai localities. Anything else shows as "area not recognised".
- Pins are at the locality, not the house.
- It reads text only — not voice notes or photos.
- It has not been used in a real flood. The results page is a simulation.
Privacy
There is no FloodLine server. Messages stay in the volunteer's browser. If Gemini is switched on, the message text is sent to Google's Gemini API to be read, and nowhere else.
Who made it
Nethra R, a student, built FloodLine in November 2026. The code is open source under the MIT licence: github.com/Nethra-R368/floodline. If you run a volunteer group and want to try it, or you see something wrong, open an issue there.