Skip to main content
TL

Agronomic trials platform

Ongoing

Taking agronomic trials out of paper notebooks and Excel, through to the dashboard.

trials per year
350
layout generation
< 1 s
valid generated layouts
100 %
Illustration of the agronomic trials platform

Context

  • Technicians record their readings by hand in the plots, then re-key them into Excel at the end of the season.
  • Each season reuses the previous season's Excel charts to analyse results.
  • Users range from technicians to farmers and work in the field as well as the office. The tool has to run on a tablet, with a network connection.

My mission

  • Building most of the backend, in NestJS and MongoDB.
  • Initial trial layout generation algorithm, under distribution constraints.
  • React frontend, from about 15 % of the scope at first to its ongoing handover.
  • Taking over the enhancements after the first release: estimation, specification clarification and client exchanges.
  • Technical documentation for the client, with the API routes and screenshots placing each call in the interface.
  • Incremental reduction of technical debt on the modules being touched.

Timeline

  1. 2026
    • Gradually handing the application over to a colleague, while remaining the reference on technical, process and domain matters.
  2. 2023
    • Developing new features, and acting as the client's technical contact on estimation and specification clarification.
  3. 2022
    • First release, and move to a permanent contract after demonstrating technical autonomy on the core of the product.
  4. 2021
    • Joining as a work-study student and building the trial layout generation algorithm.

Decisions and trade-offs

  1. 01

    Generating trial layouts from a fixed starting point rather than at random

    Problem

    Generation fills a matrix of plots under distribution constraints. The first version starts from a random column, to vary the layouts.

    Solution

    A benchmark measures under 50 % valid layouts with a random start on small sizes, the most common ones. A fixed starting point makes generation reliable on the sizes and trial types in use.

    Why this choice

    On these sizes, a valid layout matters more than variety. In exchange, layouts vary less.

  2. 02

    Aligning the existing code in small passes rather than one rewrite

    Problem

    The core module stays in technical layers, with no tests. Elsewhere, the team adopts a split inspired by Clean Architecture.

    Solution

    On each pass, I restructure one sub-module toward that split and cover it with tests.

    Why this choice

    A one-shot rewrite takes long and leaves changes untested. Short passes give complete examples to the colleague taking over the application. In exchange, the two styles coexist without converging.

What this project taught me

  • Building skills autonomously under demanding code review, on a first work-study project.
  • Validating a design assumption by benchmark, on the valid-layout rate of small matrices.
  • Taking on estimation, specification clarification and client exchanges on the enhancements after the first release.
  • Switching to the backend after a few weeks. On the frontend, review back-and-forth slowed velocity.

In hindsight

  • Spend more time on planning and overall quality before shipping. The habit of rushing to production left debt and multiplied round-trips on small bugs, in the application or in the React Query data flow.
  • Invest earlier in operations research for layout generation. The next version, in constraint programming, shows the gain of a better algorithmic base from the start.