# PrismEngine

> Self-hosted AI media infrastructure for ingest, worker processing, invisible watermarking, governed publication, and hybrid semantic search.

## Canonical URLs

- en: https://prismengine.adrienlejeune.com/
- fr: https://prismengine.adrienlejeune.com/fr/
- sitemap: https://prismengine.adrienlejeune.com/sitemap.xml
- robots: https://prismengine.adrienlejeune.com/robots.txt

## Product Context

PrismEngine orchestrates ingest, processing, protection and publication for your media on your own infrastructure. One control layer to move faster and keep ownership of the pipeline.

PrismEngine is positioned for AI infrastructure engineers, platform engineers, technical operators, and small teams that need an owned media pipeline instead of a closed SaaS workflow.

## Core Capabilities

- End-to-end flow: Ingest, validation, processing, protection and publication stay in one pipeline. Fewer breaks, fewer tools.
- Self-deployable: Run PrismEngine on your own infrastructure, with security from ingest to publication.
- Built to operate: Dashboard, automation, APIs and governance turn your library into a reliable publishing engine.
- Object storage first: Originals, previews, HLS outputs, and worker artifacts live in S3-compatible buckets so storage remains portable and CDN-ready.
- Async worker boundary: Transcoding, preview generation, watermark validation, and embedding tasks run outside the request path to keep APIs and dashboards responsive.
- Search stack: Meilisearch, structured filters, vector embeddings, and OpenCLIP-style semantic signals combine lexical precision with AI retrieval.
- Public/private split: Public APIs, private administration, and internal health routes stay separated by host, route policy, and access controls.
- Deployable runtime: Astro SSR, Docker health checks, strict security headers, canonical URLs, and crawler-readable pages keep the web layer production-ready.
- Ingest starts with qualified media: Large media files land cleanly on S3-compatible storage, with ingest state tracked before the next steps begin.
- Resumable upload: Large transfers resume cleanly instead of starting from zero.
- S3-compatible storage: Originals and derivatives live in S3-compatible storage, not a closed upload silo.
- Tracked ingest state: Ingest state stays explicit so uploads can be monitored, resumed and operated.
- Heavy processing runs on workers: Previews, HLS, transcoding and AI enrichment move through dedicated workers while the backend, admin and APIs stay responsive.
- External workers: CPU-heavy jobs run outside the server that handles the dashboard and APIs.
- Production derivatives: Photo previews, video renditions and HLS outputs are generated for real delivery workflows.
- Invisible watermarking with proof: Built on VideoSeal, PrismEngine embeds an invisible mark, checks it on key outputs and keeps proof attached to the asset.
- Invisible watermark: Embedded into pixels rather than added as a visible overlay.
- Verified detection: Detection is checked on the source and on useful derivatives before media is treated as protected.
- Proof trail: Verification state, versions and proof stay attached to the media.
- Public exposure is governed: Public is more than a visibility flag. Media must be ready and structurally consistent before galleries or APIs expose it.
- Actually public-ready: Media does not reach public surfaces unless it is genuinely ready for distribution.
- Hierarchy rules: A public media item cannot contradict the state of its album or series.
- AI search finds what keywords miss: Semantic search, visual similarity and structured filters help teams retrieve media by intent, resemblance or metadata.
- Semantic search: Visual similarity and intent matching help when the exact term is missing.
- Production filters: Date, location, media type, format, resolution and dimensions remain queryable.

## Technical Entities

- PrismEngine
- self-hosted AI media infrastructure
- AI media pipeline
- AI infrastructure engineer
- media operations
- Docker
- Docker Compose
- S3-compatible storage
- object storage
- async workers
- job queue
- Celery
- PostgreSQL
- Redis
- Meilisearch
- HLS
- transcoding
- AI enrichment
- embeddings
- semantic search
- visual similarity
- invisible watermarking
- VideoSeal
- provenance
- publication governance
- media asset management

## Public Crawl Policy

- Public marketing pages are intended to be indexed and summarized.
- The health check endpoint is operational only and should not be indexed.
- The site exposes canonical URLs, hreflang alternates, JSON-LD structured data, robots.txt, sitemap.xml, and this llms.txt guide.

## Author And Entity

- Creator: Adrien Lejeune
- Creator URL: https://adrienlejeune.com/
- Official site: https://prismengine.adrienlejeune.com/