The AlgoFly Journal.
Inside the tools we build and the ideas that move us forward.
Practical perspectives on AI, engineering, and open source.
How to Train a Custom OCR Model with Florence-2 on Limited Data
Move beyond generic OCR. Learn how to fine-tune Microsoft's powerful Florence-2 Vision Language Model on your custom dataset to extract exactly the text you need with incredible accuracy. Optical Character Recognition (OCR) has been a long-
Train Image Embeddings for Similarity Search with DINOv2 and Open Metric Learning
Fine-tune a domain-specific image representation, build a Qdrant gallery, and evaluate neighbors without confusing a custom experiment with a universal benchmark.
Building Product Price History from Millions of Retail Observations
A practical companion to AlgoFly’s retail price-history demo: product identity, time-series observations, indexing, and comparisons that keep their context.
From a Shelf Photo to Product Matches: Designing a Retail Recognition Pipeline
How to separate shelf detection, product retrieval, unknown items, and human review in a practical retail vision workflow.
A Small React Hook for Nested State: The Idea, the Pitfalls, and a Safer Version
Explore AlgoFly’s path-based React state experiment, including immutable updates, functional setters, false values, and when to choose a reducer instead.
Agricultural Field Delineation with Segment Anything and QGIS
From satellite imagery to editable field polygons: promptable segmentation, geospatial coordinates, tile joins, geometry review, and field-level information.
AI for Basketball Analysis: Tracking Players, Shots, and Identity
Inside the engineering decisions behind basketball analytics: camera placement, player reidentification, shot-event sequences, uncertainty, and useful statistics.