AI & ML pipelines templates
Model training, RAG and agent architectures. Every one opens as a real, editable diagram — restyle it, extend it and export it as your own. See also the data pipeline diagram tool.
All 14 ai & ml pipelines templates
RAG (retrieval-augmented generation) pipeline
Documents indexed, retrieved, and used to ground an LLM answer.
MermaidML model training pipeline
Data to a deployed model, with an evaluation gate.
MermaidLLM agent tool-use loop
An agent choosing between answering and calling a tool.
MermaidModel inference serving
A request path through a model server with caching.
MermaidCustomer support chatbot architecture
Intent routing between an LLM and a human handoff.
MermaidData labeling pipeline for ML
Raw samples through human review into a training set.
MermaidComputer vision inference at the edge
Cameras feeding an edge device that runs an optimised detection model locally, sending only events and samples to the cloud for retraining — for computer vision and IoT teams.
MermaidRecommendation system architecture
A two-stage recommender: user events feed a feature store, candidate generation retrieves items, a ranking model scores them and business rules re-rank — for ML and product engineers.
MermaidNLP text classification pipeline
Classifying support tickets or reviews: text cleaning, tokenisation, a fine-tuned transformer, confidence thresholds and human review for low-confidence predictions — for NLP engineers.
MermaidSpeech-to-text pipeline
Turning call or meeting audio into searchable transcripts: voice activity detection, Whisper transcription, speaker diarisation, PII redaction and summarisation — for voice AI teams.
MermaidAnomaly detection pipeline
Detecting unusual behaviour in metrics or transactions: streaming features, an unsupervised model, dynamic thresholds, alert routing and analyst feedback — for ops and risk teams.
MermaidA/B model evaluation & shadow deployment
Safely promoting a new ML model: offline evaluation, shadow traffic, canary, online A/B test with guardrail metrics and automatic rollback — for MLOps and data science teams.
MermaidFeature store architecture
An offline and online feature store: batch and streaming pipelines compute features once, a registry defines them, and training and serving read consistent values — for ML platform teams.
MermaidEmbedding & vector search pipeline
Building semantic search: documents are chunked, embedded and indexed in a vector database, then queries are embedded, matched with hybrid search and re-ranked — for search and AI engineers.
Mermaid