Architectural Deep-Dive

Krishi Dristi

An ongoing offline-first aerial intelligence prototype being developed for the realities of fragmented farms in Bangladesh.

NOTE: This is an ongoing prototype. The write-up documents the current architecture, experiments, and lessons so far rather than a finished deployed system.

Product context

Krishi Dristi is an offline-first agricultural intelligence prototype. It explores crop issue detection under fragmented-farm and low-connectivity constraints rather than assuming the large, uniform fields used by many precision-agriculture examples.

Architecture direction

The prototype combines image capture, local processing, computer vision, and ML experiments. Offline-first is a system constraint: the useful path should not depend on a reliable connection at the moment field data is collected.

Engineering lessons

Non-functional constraints shaped the architecture more than the first feature list.
Local processing reduces dependence on hosted vision services.
Plot-level reasoning is more relevant to the target context than treating every field as one uniform unit.
The project exposed the difference between a promising notebook experiment and a field-ready product.

Evidence and limits

This remains a prototype and research direction. It is not a deployed agricultural service, and the portfolio does not claim validated model accuracy, field adoption, or production readiness. The value of the project is the constraint-driven system design and the lessons from testing assumptions.