How FAISS enables face search across 1 billion photos in under a second

·1 views

Face2social founder details the vector-search engineering behind a face-recognition engine capable of scanning roughly one billion photos in under a second. Each uploaded face is converted into a high-dimensional vector, turning facial similarity into a geometric nearest-neighbor problem. Brute-force exact search becomes physically impossible at billion-scale due to memory and bandwidth constraints, not processing power alone. Facebook's open-source FAISS library solves this by using approximate nearest-neighbor indexing, trading perfect recall for dramatic speed and memory gains. The core engineering challenge, the author explains, lies in balancing the three-way trade-off between search speed, memory usage, and result accuracy.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How Quarkus and Micronaut Apps Can Share Messages via RabbitMQ

Modern distributed systems often combine services built on different JVM frameworks, creating a need for seamless inter-service communication. A technical guide demonstrates how RabbitMQ can bridge a Quarkus application and a Micronaut application using asynchronous messaging over the AMQP protocol. The setup involves a Quarkus service that publishes LeaveRequest messages and a Micronaut service that consumes and processes them. A shared Gradle module named 'common' holds the LeaveRequest data model, ensuring both applications operate on a consistent message contract. The integration relies on SmallRye Reactive Messaging's RabbitMQ connector on the Quarkus side and Micronaut Serde on the consumer side to handle serialization and channel routing.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Open-source directory lists 50+ free public APIs with daily automated health checks

A developer has launched Awesome Free APIs Live, an open-source curated directory of over 50 free public APIs spanning categories such as AI, developer tools, security, and open data. The project addresses a common frustration among developers: API directories that list services which have since gone offline or switched to paid models. Using GitHub Actions, an automated script runs daily health checks on all listed endpoints and updates live status badges accordingly. Broken or inactive endpoints are flagged automatically, saving developers time spent debugging non-functional services. The directory is hosted on GitHub Pages and is open to community contributions via pull requests on its public repository.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Google's August 2026 Spam Update Caused 82% More Top-10 Ranking Drops Than Baseline

Google rolled out its third spam update of 2026 between August 18 and August 21, completing the process in just over three days, as recorded on its official Search Status Dashboard. Unlike a major policy overhaul, the update enforced existing spam guidelines but still triggered significant ranking shifts for many websites. Data from SE Ranking, cited by Search Engine Land, showed that 16.71% of URLs previously ranking in the Top 10 fell beyond position 100 during the update, compared to 9.2% in the July baseline — roughly 82% higher volatility. A drop from Top 10 to beyond position 100 can effectively eliminate a page's search visibility, directly cutting organic traffic and leads. SEO experts caution that while the update coincided with these losses, site owners should conduct careful diagnosis rather than assume a penalty, as factors like thin AI-generated content, competitor changes, or technical issues may also be at play.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Engrava Posts Reproducible LongMemEval-S Benchmark Scores for Versions 0.5.0 and 0.6.0

Memory system Engrava published verifiable benchmark results on the 500-question LongMemEval-S test, with version 0.6.0 scoring 81.6% micro in August 2026 and version 0.5.0 scoring 82.4% micro in July 2026. Both runs used the same canonical scorer, standard GPT-4o reader and judge, and a top-k of 20 retrieved turns, with no changes to anything outside the memory layer itself. Unlike many published benchmarks, Engrava released full reproduction artifacts for both runs, including the older, slightly higher-scoring result that predates the current release. The memory pipeline contains no generative language model — ingestion and retrieval rely on deterministic hybrid search over a typed graph, meaning no LLM tokens are consumed on every read or write operation. The four-question gap between the two versions prompted the team to examine run artifacts directly rather than attributing the difference to either regression or statistical noise.

0 comments Read more at DEV Community