# Turning drives full of field photos into a searchable archive

Page: https://pixelteh.com/work/agritech-image-archive · Last updated: 2026-11-16

_Agritech · Image catalogue · 2022 to 2023 and 2025_

An agritech startup helps farming companies organise their aerial and field photography. We built the platform that replaced hard drives and memory cards, and the client came back two years later to add AI.

## Challenge

Farming companies kept aerial and field photos on drives and memory cards. Finding an older image took hours, if it was possible at all.

## Outcome

A platform for resumable batch uploads, custom categorisation and search, later extended with AI tagging. Finding an image takes seconds and 27 companies use it.

## Key results

- **Seconds**: to find an image, instead of hours
- **27**: companies on the platform
- **500**: images per upload batch, up to 60 MB each
- **1000s**: of images per company

## Facts

- **Client**: Agritech startup
- **Engagement**: 2022 to 2023, plus a 6-month AI phase in 2025
- **Team**: 2 developers
- **Status**: Completed

## The situation

The startup's customers kept their images on shared drives, hard drives and memory cards. Finding an older photo took hours and sometimes wasn't possible at all. Because the archive was so hard to search and share, it produced almost no new knowledge.

## What we built

- Resumable batch uploads of 400 to 500 images of up to 60 MB each, with retry, pause and resume, through AWS multipart upload.
- A background pipeline that reads metadata, generates derived files and applies each customer's own categorisation rules.
- Search and filters by project, location, date, asset type, category and custom fields.
- Annotations, discussion threads and notes stored and searchable alongside the images.
- Rate limiting, backpressure, retries and dead-letter queues so large uploads never slow down the rest of the system.
- In 2025, AI classification and tagging added to the processing pipeline.

## Results

Finding an image went from hours, or never, to seconds or minutes depending on how deep the search goes. Upload speed itself didn't change much. Behaviour did: companies started uploading their photos instead of keeping them on cards and drives. By the end of the first engagement 27 companies used the platform, each with thousands of images, and their teams spotted problems and opportunities in the field sooner.

## How the engagement evolved

The platform was completed in 2023 without AI. In 2025 the client came back for a six-month engagement to add AI classification and tagging.

## What the client said

> The finished product was the first major achievement of our business. Pixelteh's development work significantly impacted the business. The platform is now building traction. Their technical expertise and positive attitude were noteworthy traits of the engagement. Customers can expect a communicative and accessible development team.
>
> Adriana, COO, agritech startup

## Stack

Node.js, PostgreSQL, AWS S3 (multipart upload), RabbitMQ, Redis, React, TypeScript, Tailwind CSS, AI classification

## Related

- [Build a new product](https://pixelteh.com/services/product-development)
- [Add AI to an existing system](https://pixelteh.com/services/ai-integration)
- [Operations and field services](https://pixelteh.com/industries/operations-field-services)
- Next case study: [From WhatsApp groups to a staffing platform that runs every event](https://pixelteh.com/work/event-staffing-platform): Worker no-shows from 11% to 4% for a US event-staffing agency
