All work

04 · Product

SemaiSens

Agri-tech software that tells Malaysian growers what to do next and shows the evidence behind each alert.

Role
Product lead, end to end
Context
Growers, estates and agencies in Malaysia
When
2026
Focus
Agri-tech, Data product, Explainable AI, SaaS
4crop models built for tropical farming
8stages from satellite pass to decision
10 mfree satellite imagery on every plan
3paid plans, from smallholder to estate
From satellite pass to a decision you can check
  1. 01Satellite pass, with cloudy passes marked
  2. 02Indices matched to the crop and growth stage
  3. 03Rule cards based on published research
  4. 04Semai Advisor, which can decline to answer
  5. 05Alert with its evidence
  6. 06Field action logged as a record

The problem

Plenty of farm-monitoring tools can show a green map. Few tell a grower what to do on Monday morning, and almost none can show afterwards why the advice was right.

Malaysian farming also doesn’t fit the usual models, which are mostly built for temperate row crops. The crops that matter here are flooded paddy, long-lived oil palm, rubber tapped on rotation and pineapple planted in batches. Growers, estate managers and agencies all want different answers from the same data.

My role

I led the product end to end: positioning, the decision pipeline, the information architecture for each role, the grower app and the public site.

Constraints

  • Satellite passes come every 3 to 7 days, and many are cloudy.
  • Advice has to be explainable. A wrong call can cost a harvest, so “the AI said so” won’t do.
  • The same product has to work for a smallholder on a phone, a field team and an estate manager.
  • The free satellite data has to be useful by itself, so growers can start without paying.

Decisions

Evidence behind every alert

The positioning line is “Everyone can show you a green map. We can prove what happened on it.” Each alert links back to the field, weather and satellite signals behind it, so a team can see what changed and why it matters.

Crop models built for the tropics

There are four crop models, for paddy, oil palm, rubber and pineapple, each designed around that crop from the start. Indices are matched to the crop and its growth stage.

A pipeline that checks itself

A satellite pass goes through eight stages, in order, before it becomes a decision, and two of those stages exist only to catch the system being wrong. A cloudy pass shows up as hollow on the timeline instead of being skipped. Advice comes from rule cards tied to published research, and the advisor declines to answer when the evidence is thin.

One view per role

Growers see what to do next, field teams see their priorities, managers see the whole operation, and agencies compare regions. Every view runs on the same data.

Free to start

Sentinel-2 imagery at 10 m resolution is free on every plan, along with indices, alerts and the advisor. Paid plans add sharper and more frequent imagery, tasks, field-team coordination, variable-rate maps and regional forecasts.

In the grower’s language

Notifications and analysis reports go out in Bahasa Malaysia, for example “Ladang anda telah dianalisis” (your farm has been analysed).

Outcome

  • Live as a public product, with a free plan plus Grower and Estate plans.
  • Four tropical crop models, more than 100 crop guides, and an eight-stage pipeline that checks its own output.

What I’d do next

  • Measure adoption properly: weekly active fields, alerts acted on, and the time from an alert to a field visit.
  • Close the ground-truth loop, so field observations recalibrate the models.
  • Build the verification bundles planned for subsidy, takaful and disaster-relief partners.