AI/ML
AI/ML recipes
8 admin-reviewed AI/ML starting points in the Vault, each installable with one command. Browse every category →
AI Content Classification & Moderation
Auto-classification via an LLM (toxicity, spam, category tagging) with a worker to process items async and Redis to avoid re-classifying the same content twice. Distinct from the Content Moderation Pipeline recipe: that's the human-review queue; this is the AI doing the first pass.
by @ops
Go + Redis: Distributed Feature Store for ML Model Serving
Low-latency feature lookups for ML models at inference time — a real MLOps pattern distinct from anything else in the catalog's AI/ML section: this doesn't call an LLM or process embeddings, it serves pre-computed features fast enough for real-time model inference.
by @ops
Podcast/Video Transcription & Captioning Service
Audio/video files land in MinIO, a Celery worker runs speech-to-text and generates captions/transcripts stored in Postgres. Distinct from AI Content Classification: that classifies existing text, this creates text from audio in the first place.
by @ops
Python + pgvector + MinIO: Duplicate Image/Media Detection Backend
Perceptual-hash and embeddings-based detection of duplicate or near-duplicate images, a worker processing uploads and comparing against stored vectors — a distinct application of pgvector from document search or recommendations: finding near-identical media, not similar meaning.
by @ops
Python + pgvector + Redis: AI SaaS Starter
A production-ready backend for LLM-powered products: FastAPI-style Python API, Postgres with pgvector for embeddings and semantic search, Redis for response caching and rate limiting, and a background worker for async completions and embedding jobs.
by @ops
Python + pgvector + Redis: Product Recommendation Engine
Embeddings-based product similarity and recommendations, a worker computing recommendation sets on a schedule — distinct from Semantic Document Search: that finds documents by meaning, this recommends catalog items by similarity, an e-commerce-shaped application of the same underlying technique.
by @ops
Python + pgvector: Semantic Document Search
Embeddings-backed search over a document corpus — distinct from the AI SaaS Starter (that's a RAG chat product) and Search-as-a-Service (Meilisearch keyword search): this is specifically finding documents by meaning, not exact terms.
by @ops
Python + Postgres + MinIO: Text-to-Speech (TTS) Synthesis Pipeline
Generates audio from text via a worker, storing output in MinIO — the reverse direction from the Transcription/Captioning recipe: that turns audio into text, this turns text into audio.
by @ops