AI-Based early warning and landslide Risk Monitoring System in NER

The receipts 🧾

Every red flag on this statement, in full. These are the four places it bites.

1. Exhibit A

   The description lists soil moisture sensors as an input and there is no such deployed network you can read — teams either quietly drop it or fake it, and a judge who notices will treat the whole input stack as suspect

2. It gets worse

   Landslide early warning is heavily attempted every SIH cycle; you will be one of many, and the marginal team is separated by validation rigour, not by dashboard polish

3. Still reading?

   Offline plus multilingual plus SMS is a full day of unglamorous plumbing that teams always schedule last and always run out of time for

4. And the finisher

   Without a stated lead time you have no target — predicting a landslide six hours out and six days out are different problems, and picking the wrong one makes your evaluation meaningless

The damage report 📉

Every score this statement earned, and what each one actually costs you.

• Feasibility

  3/5

  Buildable. Not comfortably. There is a week in here you have not planned for yet.

  DEM, slope and satellite layers are free from Bhuvan and SRTM/Copernicus and GSI's landslide inventory gives you labels, but the description's soil-moisture sensor network does not exist for you to read from and NER landslide records are sparse and coarsely geolocated, so your training labels will be thin where it matters most.

• Innovation scope

  2/5

  Nothing here is new. Your only edge is execution — and execution is also everyone else's only edge.

  Points a through f fix the data sources, the alert recipients, the GIS layer, the citizen upload channel and all four dashboard panels, so what is left to you is model choice and nothing about the product shape.

• Clarity

  4/5

  The ask is unambiguous, which quietly removes your favourite excuse.

  The deliverables are enumerated tightly down to the four dashboard panels and the integration targets, but the description never says what counts as a correct prediction — no lead time, no spatial resolution, no accuracy floor — which is exactly the thing a judge will press you on.

• Acceptance potential

  3/5

  Middle of the pack. This statement will not win the room for you — you will have to.

  Solid and genuinely wanted, but landslide and flood early-warning is one of the most repeatedly attempted themes in SIH history, and with no linked dataset your accuracy claim rests on labels you assembled yourself.

• Effort

  Heavy

  Heavy. Somebody on this team is not sleeping in week three. Pick who, on purpose.

  Multi-source ingestion, a trained risk model, a GIS dashboard, a separate field-reporting mobile app, an SMS gateway, multilingual content and offline sync are six distinct workstreams, and the ML is the smallest of them.

• Demo-ability

  Medium

  Demoable, if you rehearse it. Nobody rehearses it.

  You cannot produce a landslide, so the whole demo rests on a historical replay being convincing; it works, but only if you have picked a well-documented event and can show the model was not simply trained on that day.

• Data

  None supplied

  No dataset comes with this one, so every accuracy figure you quote is a number about labels you invented.

  Nothing is provided with the statement. You are sourcing, cleaning and labelling it yourself, and that work is invisible in the demo but very visible in the questions.

The demo they will have already seen 🪞

Somewhere around 140–330 teams are heading here, and the description is doing the choosing for most of them. They will read the same brief, reach the same architecture, and build a version of the same demo you are planning. Being correct is the floor. If your five minutes could be swapped with the team before you and nobody in the room would notice, you have not picked badly — you have built predictably, which costs exactly the same and hurts more.

What survives 🛡️

The ground worth standing on when the questions start.

• Landslide susceptibility mapping is a mature published field, so you can anchor your feature set in accepted literature rather than inventing predictors and defending them from scratch
• The citizen geo-tagged upload requirement gives you a second, cheap demo surface that works even when the model is not convincing
• MDoNER as a sponsor means the evaluation panel will know NER terrain specifically, so naming real corridors like the Imphal–Jiribam route buys credibility that a generic map does not