About

Project Overview We operate a commercial vegetable farming operation and are exploring the use of computer vision and data-driven modeling to improve harvest forecasting accuracy. We are seeking an experienced engineer with strong computer vision experience and sound ML judgement to help design and deliver a proof-of-concept (POC). The goal of this phase is not to build a full production system, but to determine what is feasible, what is reliable, and where machine learning genuinely adds value versus simple logic. This project is intentionally pragmatic. We are not looking for ML hype or over-engineering. --- Problem Description We want to assess whether harvest readiness can be predicted using a combination of image-based measurements and historical growth data. At a high level, the system would: * Measure current crop size from smartphone images taken in the field * Use a physical reference object in each image to infer scale * Track size distributions across multiple heads per image (typically ~20–30) * Combine image-derived measurements with historical growth data and weather information * Estimate when the crop is likely to reach harvestable size, with appropriate uncertainty The expectation is to start with simple, explainable approaches and introduce ML only where simpler methods are insufficient. --- Scope of the Proof of Concept This engagement is deliberately limited in scope. The POC should help answer: * Can crop heads be detected and measured reliably in real-world outdoor conditions? * What computer vision approach is simplest and robust enough for this task? * How accurate can size estimation be using smartphone images and a reference object? * When do rule-based growth models work, and where do they break down? * Where does machine learning materially improve predictions? * What would a scalable future system require in terms of data, tooling, and effort? You do not need to build a polished app, production pipeline, or deployment infrastructure. --- What We Are Looking For We are specifically looking for someone who can guide the technical approach, not just implement instructions. Ideal experience includes: * Computer vision for object detection and instance segmentation * Practical experience measuring real-world objects from images * Strong understanding of when ML is appropriate versus rule-based logic * Designing and evaluating proof-of-concepts * Clear communication of trade-offs, assumptions, and risks Agricultural experience is a plus but not required. --- Deliverables (POC-Level) Expected outcomes may include: * A prototype or notebook demonstrating head detection and size estimation * Evaluation of accuracy, failure cases, and environmental sensitivity * A simple growth modeling approach with clear limitations * Recommendations on whether and how to proceed to a larger build Deliverables can be refined collaboratively. --- Engagement Details * Initial engagement is for a proof of concept only * There is potential for follow-on work if the POC is successful * Please include relevant examples of similar work or describe how you would approach this problem --- Screening Questions (Required) Please answer the following in your proposal: * How would you approach this problem at a POC level, and what would you deliberately avoid over-engineering? * What computer vision techniques would you consider for measuring multiple crop heads in outdoor conditions? * Where would you expect logic to work, and where would you expect ML to be necessary? * What do you see as the biggest technical risks or unknowns in this approach?