AI Tools & Infrastructure Navigator
What component of an AI system must I choose, connect, or operate?
Choose the branch that matches your decision, then choose the relevant article. The destinations are alternatives and references, not a mandatory reading sequence.
Start: Choose Your Decision
- Do you need an end-user assistant or a model-facing service? → Assistants and hosted model families
- Are you implementing a model, an application, or an analysis workflow? → Libraries and programming tools
- Where must computation run? → Local, cloud and hardware environments
- What must communicate with the AI component? → Integration and data services
- What prevents the current model from meeting requirements? → Model adaptation and compression
- What changes when the experiment becomes a service? → Deployment and lifecycle tooling
- Does the task need a document, speech, vision, creative, or domain-specific service? → Specialized capabilities
- Do you need a fast experiment, notebook, or model repository? → Prototypes and research workspaces
If you are unsure, begin with the first branch and state the outcome you need before selecting a method or tool.
Assistants and hosted model families
Question: Do you need an end-user assistant or a model-facing service?
For interactive work, compare assistants on your own tasks. For application integration, check the model interface, access terms and operational requirements.
Article destinations: Narrow the choice: Which group matches your input or development environment?
Models, providers and model families
- Alibaba
- Alibaba Qwen
- Alpaca
- Apple
- BLIP-2
- Baidu
- Cohere
- DeepSeek
- ERNIE
- Ernie
- FLAN-T5 LLM
- Gato
- ImageBind
- Kosmos-1
- LLaMA
- Meta
- Meta Muse
- Mistral
- PaLM
- RETRO
- Sakana
- Stability AI
- T5
- Toolformer
- Wolfram
- XAI
- XLNet
Interactive assistants and companion experiences
- Anthropic Claude Cowork
- Apple Siri AI
- Character.AI
- DeepSeek Harness
- Gemini Notebook
- Google Gemini Spark
- Grok
- Grok Bot
- Microsoft Copilot Autopilot
- OpenAI dots
- Perplexity Computer
- Phind
- Pi
- PoE (Platform for Open Exploration)
- Replika
- Silly Tavern
- You
- You Research Agent
Checkpoint: Test the same task on candidate systems; verify current availability and terms.
- Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
- Not yet: Revisit the relevant explanation or choose a simpler baseline.
- Choose another decision →
Libraries and programming tools
Question: Are you implementing a model, an application, or an analysis workflow?
Choose the language and deployment environment first, then compare compatible libraries. Existing library articles may describe historical tools; verify maintenance before adoption.
Article destinations: Narrow the choice: Which group matches your input or development environment?
- Machine-learning libraries and platforms →
- Language-specific and desktop programming →
- Experimentation, serving and lifecycle tools →
- JavaScript and browser tools →
- Language, retrieval and knowledge tools →
Machine-learning libraries and platforms
- AI Hub — Google AI Hub / TensorFlow Hub context; this is not the PRIMO.ai navigation hub.
- AdaNet
- AmazonML
- Anaconda
- Apache Mahout
- Auto Keras
- Azure AI Process
- Azure Machine Learning Studio
- Baidu Qianfan Agent Development Platform
- BiqQuery ML (BQML) — existing title retained; review the apparent BigQuery spelling error.
- Caffe / Caffe2
- Cloudera
- Coding TensorFlow
- Cognitive Services
- Colaboratory
- DataRobot
- Databricks
- Differentiable Programming
- Git - GitHub and GitLab
- Gluon
- Google Antigravity
- Google AutoML
- Google Semantic Reactor
- H20 — overlapping H2O coverage; canonical identity needs editorial review.
- H2O — overlapping H2O coverage; canonical identity needs editorial review.
- Intel® Compute Library for Deep Neural Networks (clDNN)
- Kaggle
- Kaggle Kernels
- Keras
- Libraries & Frameworks
- Libraries & Frameworks Overview
- Ludwig
- ML Engine
- MXNet
- Manifold - Uber
- Neon
- OpenMined
- Other Coding options
- Palantir
- Prediction API
- Project Cortex
- RAPIDS
- Repositories & Other Algorithms
- SageMaker
- Shogun
- Spark MLlib
- TFLearn
- Theano
- Turi
- Vertex AI
- Weka
- Zapier
Language-specific and desktop programming
- Accord.Net Framework
- Android Development
- Cython
- Deeplearning4j
- GoLearn
- Matlab
- Microsoft Cognitive Toolkit
- Neuroph
- R Project
- Swift
Experimentation, serving and lifecycle tools
- Aviary — Anyscale model-serving tooling.
- Dopamine
- Eggplant
- Google Cloud Code
- Humanloop
- Ray - UC Berkeley RISELab
- Replit
- Spell
- TensorBoard
- TensorWatch
JavaScript and browser tools
Language, retrieval and knowledge tools
- DialogFlow
- Gensim
- Google Cloud Natural Language API
- Google Natural Language
- Lex
- Natural Language Toolkit (NLTK)
- Natural Language Tools & Services
- Sintelix
- SpaCy
Checkpoint: Build a minimal example with the intended runtime and dependency versions.
- Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
- Not yet: Revisit the relevant explanation or choose a simpler baseline.
- Choose another decision →
Local, cloud and hardware environments
Question: Where must computation run?
Use local or edge environments when control, connectivity or latency requires it. Compare hosted infrastructure when managed operations or scale matter.
Article destinations:
- AWS Internet of Things (IoT)
- AWS IoT Button
- AWS with TensorFlow
- Azure with TensorFlow
- Bedrock
- Deep Learning (DL) Amazon Machine Image (AMI) - DLAMI
- DynamoDB
- Google Coral
- Google Internet of Things (IoT)
- Greengrass
- Groq
- IBM
- Intel
- Kinesis
- Lambda
- Laptop
- Local — broad or mixed title; inspect the article context before proceeding.
- Management Console
- Microsoft Internet of Things (IoT)
- Neural Compute Stick (NCS)
- Pixel — Google Pixel and Tensor hardware.
- Platforms: AI/Machine Learning as a Service (AIaaS/MLaaS)
- Power (Management)
- Simple Notification Service (SNS)
- Simple Queue Service (SQS)
- Simple Storage Service (S3)
- SingularityNET
Checkpoint: Verify memory, acceleration, connectivity, access and cost constraints.
- Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
- Not yet: Revisit the relevant explanation or choose a simpler baseline.
- Choose another decision →
Integration and data services
Question: What must communicate with the AI component?
Identify documents, databases, APIs, events and message queues. Choose the interface and failure behavior before a product.
Article destinations:
- Architectures
- Data Flow Graph (DFG)
- Directed Acyclic Graph (DAG)
- Google Sheets
- MQTT
- Messaging & Routing
- Wikifier
Checkpoint: Demonstrate a complete request, response and failure-recovery path.
- Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
- Not yet: Revisit the relevant explanation or choose a simpler baseline.
- Choose another decision →
Model adaptation and compression
Question: What prevents the current model from meeting requirements?
Improve prompts or retrieval for missing context; consider transfer or fine-tuning for specialized behavior; examine compression for runtime constraints.
Article destinations:
- Converting to TensorFlow Lite
- Image/Video Transfer Learning
- Neural Network Pruning
- Prompting vs AI Model Fine-Tuning vs AI Embeddings
- Text Transfer Learning
- Train Large Language Model (LLM) From Scratch
- Transfer Learning With Keras
Checkpoint: Measure the effect on both task quality and operating requirements.
- Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
- Not yet: Revisit the relevant explanation or choose a simpler baseline.
- Choose another decision →
Deployment and lifecycle tooling
Question: What changes when the experiment becomes a service?
Plan packaging, serving, versioning, monitoring and rollback. Distinguish a model-serving tool from a full operating process.
Article destinations:
- AWS Lambda & Python
- Kubeflow Pipelines
- ML Test Score
- Serverless
- TensorFlow Extended (TFX)
- TensorFlow Serving
Checkpoint: Show how a failed release can be detected and rolled back.
- Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
- Not yet: Revisit the relevant explanation or choose a simpler baseline.
- Choose another decision →
Specialized capabilities
Question: Does the task need a document, speech, vision, creative, or domain-specific service?
Compare a specialized tool with a general-purpose baseline. Avoid assuming a historical product description establishes present availability.
Article destinations:
- Afiniti
- DeepLens - deep learning enabled video camera
- Google AIY Projects Program
- Magenta
- More DeepLens Projects
- Polly
- Rekognition
- SharePoint
- SoundHound
- Spotify
- Textract
Checkpoint: Confirm that specialization improves the chosen workload.
- Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
- Not yet: Revisit the relevant explanation or choose a simpler baseline.
- Choose another decision →
Prototypes and research workspaces
Question: Do you need a fast experiment, notebook, or model repository?
Start with a small reproducible example. Use a repository or notebook environment to compare alternatives before building a production system.
Article destinations:
Checkpoint: Preserve inputs, versions and results so another person can reproduce the experiment.
- Yes, I can do this: Compare an alternative, test the choice, or continue through the related trees below.
- Not yet: Revisit the relevant explanation or choose a simpler baseline.
- Choose another decision →