HN Showcase
Back to browse
GitHub Repository

Photon is a perception engine for image pipelines: ingest, analyze, extract, and semantically index visual data. Built for high-volume processing and designed to serve as the core visual intelligence layer within larger data mining systems.

10 starsRust

Photon – Rust pipeline that embeds/tags/hashes images locally w SigLIP

by pgbouncer·Mar 2, 2026·3 points·1 comment
Visit ProjectView on HN

AI Analysis

●●SolidWizardryNiche GemShip It

Local SigLIP embeddings + 68K-term semantic tagging in a single Rust binary, zero cloud.

Strengths
  • •Single statically-linked binary eliminates runtime/dependency friction for image processing workflows.
  • •SigLIP + ONNX enables genuine offline operation with competitive semantic search quality.
  • •Batch processing with parallel workers and comprehensive output (embeddings, tags, EXIF, hashes) addresses real data pipeline needs.
Weaknesses
  • •68K-term vocabulary is opaque—no way to customize tags or understand scoring beyond 'SigLIP scored them.'
  • •Niche use case: mostly appeals to researchers/engineers already building image indexing; not a consumer or general-purpose tool.
Target Audience

Data engineers, ML researchers, and developers building image indexing or search systems.

Similar To

Clip-as-a-Service · Milvus (vector DB) · OpenSearch with vector plugin

Post Description

Open-source Rust-based image processing pipeline that takes images and outputs structured JSON — 768-dim vector embeddings, semantic tags from a 68K-term vocabulary, EXIF metadata, content hashes, and thumbnails.

Everything runs locally via SigLIP + ONNX Runtime. Single binary, no Python, no Docker, no cloud dependency. Optional BYOK LLM descriptions (Ollama, Anthropic, OpenAI).

Similar Projects

Productivity●●Solid

Clipthesis – free, local app to tag and search video across your drives

The neat trick here is indexing and deduplicating footage while still letting you search tags and preview clips when drives are disconnected — that UX quirk alone solves a real pain for solo editors. Hover-scrub thumbnails, AND-tag searches, and import helpers (select new only) show the author actually used this workflow, but it's not reinventing DAM software — it's a focused, locally-run tool for a specific audience. Mac-only and no team/cloud features limit its scope, but for one-person workflows this hits the right marks.

Niche GemSolve My Problem
hugorut
103mo ago
Developer Tools●●Solid

Snowflake Emulator in Rust – Test locally without cloud credits

This actually implements a surprisingly large slice of Snowflake: VARIANT/ARRAY/OBJECT types, LATERAL FLATTEN, QUALIFY, MERGE, COPY, transactions and ~90 JSON/array/window functions — all exposed via Snowflake SQL API v2 and compatible with the Go driver so most clients can point at localhost. Running on DataFusion in Rust and shipped as a Docker image with testcontainers/GHA examples makes it immediately useful in CI; just remember it's an emulator, not Snowflake's full runtime, so you may hit edge-case behavioral differences.

Solve My ProblemWizardryNiche Gem
sivchari
203mo ago