WeatherGPT: Conversational AI for Weather Forecasting, Alerts, and Climate Information
Ministry of Earth Sciences (MoES) · Disaster Management · Software
Buildable and demoable but entering the most crowded category on the portal against an official app that already exists — the only defensible angle is verbatim warning handling and honest uncertainty, so if you take it, make that the whole pitch rather than the conversation.
What it actually is
Weather information sits across many portals, bulletins and satellite products, and a farmer or a district officer who needs one specific answer has to know where to look. The ask is a chatbot that answers weather questions in plain language, in Indian languages and by voice, pulling from forecast models and warning systems.
What to build
A conversational weather assistant with a query understanding layer resolving a natural language question into location, time window and variable, a retrieval layer pulling live observations, numerical forecast fields and active warnings for that resolution, an answer composer that states the forecast with its issue time and uncertainty rather than a bare number and that surfaces any active warning ahead of the routine answer, multilingual and voice interaction for users who will not type, and an alert path that pushes extreme weather warnings rather than waiting to be asked, all behind a mobile-first interface.
Smallest thing that wins the room
Ask by voice in an Indian language whether it will rain over a named village in the next two days, get an answer with the forecast issue time and confidence, and have an active cyclone warning for that district interrupt the routine answer.
How crowded this one gets
A guess, projected from the 2025 statements — the last year where both the submission counts and the winners were published.
Quieter than 45% of the 226 · #124 of 226 by expected field
A normal-sized field. Your idea has to be good, not miraculous.
Why: central ministry statements sat below the average.
This is a guess, not a fact
Nobody has published 2026’s numbers yet. This is an analysed estimate from last year’s pattern, so please do not take it as the truth — check the live counter on the SIH portal before you decide anything. The range covers the middle half of likely outcomes, so one statement in two lands outside it. Entry closes at 500 ideas per statement, so no range goes past that — a statement that reaches the cap fills and shuts rather than drawing an unlimited crowd. The model reads only three things a team can see before choosing — software or hardware, the theme, and what kind of body posted it — and those explain about a quarter of the variation in last year’s field sizes (R² 0.25 on held-out statements). Trust the band more than the number, and the ordering more than either. It cannot see how good your idea is, which is the part that actually decides it.
The scores
The number is the shorthand. The line under it is the reason.
Acceptance potential
2/5A language model wrapped around a weather API is the single most common submission shape of this era, the meteorological department already ships its own app, and general assistants already answer weather questions well — so you are entering the most crowded category on the portal with nothing structural to distinguish you.
Feasibility
4/5The weather data is genuinely accessible — the meteorological department publishes observations and warnings, global forecast model output is freely available at usable resolution, and language understanding, translation and speech for Indian languages are all available through national infrastructure and open models.
Innovation scope
2/5Eight features are enumerated, the outcomes and use cases are listed, the evaluation parameters are given and even the technology stack down to specific databases and orchestration tools is suggested, leaving almost nothing about the product for you to decide.
Clarity
4/5The features, expected outcomes, use cases and evaluation parameters are all set out explicitly, but it is a feature list rather than a specification — nothing says what forecast horizon, spatial resolution or accuracy the answers must have, which is the substance of a weather product.
Effort
HeavyQuery understanding, live data integration across observations, model output and warnings, answer composition, multilingual and voice handling, push alerting and a mobile client is six components, though each is well-supported and none is technically deep.
Demo-ability
EasySpeaking a question in an Indian language and getting a spoken answer works reliably and is instantly understood, and the warning interrupt is a good moment — the problem is not that the demo fails but that every judge has seen this demo many times.
In its favour
- Green flag: Weather answers are verifiable — a judge can check your forecast against the official bulletin on their own phone, which is a stronger position than most conversational submissions have
- Green flag: The warning path is the genuinely valuable half and the half everyone will underbuild: a system that interrupts a routine forecast question with an active cyclone warning for that district is doing something a general assistant does not
- Green flag: Voice in Indian languages for users who cannot type is a real accessibility argument backed by free national language infrastructure
- Green flag: The evaluation parameters are listed in the statement, so you know response latency and multilingual capability will be assessed and can prepare measurements for both
Against it
- Red flag: This is the most cloned software shape of the moment and you will be one of many chatbot submissions in the room, several of them also over weather data
- Red flag: The meteorological department already publishes its own public app and general-purpose assistants answer weather queries competently, so the novelty bar is genuinely high and needs a direct answer
- Red flag: A confidently wrong weather answer during a cyclone is a safety issue, not a quality issue, so the system must be able to say it does not know and must never paraphrase a warning into something softer than the official text
- Red flag: No forecast horizon, resolution or accuracy requirement is stated, so a fluent answer over a coarse global model can look identical to a genuinely useful one and you should show your data provenance rather than hiding it behind the conversation
What you will be writing
- intent and slot extraction for location, time and variable
- RAG over IMD bulletins and warning texts
- GFS or WRF gridded forecast field lookup
- Bhashini multilingual and speech interface
- push alerting against active warning feeds
- FastAPI backend with real-time ingestion
- Conversational AI
- Meteorological services
- Disaster warning dissemination
Prior art to read before you start
natural language weather query answering · multilingual voice assistant for public services · extreme weather alert dissemination
Analysed by Claude Opus. Every score above is a judgment call with its reasoning attached — kindly cross-check this against the official statement on the SIH portal before your team commits to it.