AI Technology Selector
Choose AI technology based on what it must accomplish, where it must run, and how it will be maintained.
The AI Technology Selector guides comparisons among AI assistants, models, providers, frameworks, and infrastructure.
Product capabilities, pricing, availability, and terms change frequently. Treat provider links as starting points for current evaluation, not permanent rankings.
Explore Specific Tools and Infrastructure
What component of an AI system must I choose, connect, or operate?
Start with AI Tools & Infrastructure Navigator, or choose a specific branch:
- Compare assistants and hosted model families
- Explore libraries and programming tools
- Compare local, cloud and hardware environments
- Investigate specialized capabilities
- Choose a prototype or research workspace
Start: What Are You Choosing?
| Technology Decision | Continue |
|---|---|
| An AI assistant or end-user application. | A. AI Assistants |
| A model for an application. | B. AI Models |
| A development framework or service. | C. Development Tools |
| A computing or deployment platform. | D. Infrastructure |
| Technology for a specialized capability. | E. Specialized Technology |
| A comparison of competing candidates. | F. Final Selection |
A. AI Assistants
Question: What do you primarily want the assistant to do?
- General conversation, analysis, or writing: ChatGPT · Claude · Gemini
- Work within office and productivity environments: Bing/Copilot · Google · Microsoft
- Research and information discovery: Perplexity · Semantic Search
- Programming assistance: AI Pair Programming Tools
- Personal companionship: Personal Companions
- Task execution and automation: Agents/Assistants · Personal Productivity
Next question: What matters most?
- Accuracy and task success: Compare candidates using representative tasks and Evaluation.
- Privacy and data handling: Review Privacy and AI Governance.
- Integration with existing software: Explore Service Capabilities.
- Cost and operational constraints: Record current prices, usage limits, and terms for each candidate.
Checkpoint: Have you tested at least two suitable options on the same real tasks?
B. AI Models
Question: What type of model do you need?
- Language and conversation: Large Language Model (LLM) → Foundation Models (FM)
- Images or video: Vision → Video/Image
- Speech and audio: End-to-End Speech → Speech Recognition
- Structured prediction: Classification → Regression
- Generation of new content: Generative AI → Diffusion
- Recommendations or anomaly detection: Recommendation → Anomaly Detection
Next question: How will you obtain the model?
- Use an existing hosted model: AI Platform → OpenAI · Anthropic · Google
- Select a downloadable or open-weight model: Hugging Face → Model Search
- Customize an existing model: Fine-tuning → Transfer Learning
- Develop and train a model: PyTorch → TensorFlow → Train, Validate, and Test
Next question: What tradeoffs matter?
- Quality on the intended task.
- Latency and throughput.
- Cost per request or training run.
- Context size and supported modalities.
- Model licensing and access terms.
- Data handling and deployment restrictions.
- Stability and lifecycle support.
Continue: Compare candidates →
C. Development Tools
Question: What kind of development experience do you need?
- Minimal coding: Codeless Options, Code Generators, Drag n' Drop → Google AI Studio
- Python machine learning: Python → PyTorch → TensorFlow
- Web and JavaScript development: JavaScript → TensorFlow.js
- Experimentation and notebooks: Jupyter → Notebooks
- Retrieval and knowledge systems: LlamaIndex → LangChain
- Model lifecycle operations: MLflow → Algorithm Administration
- Model discovery: Hugging Face → Model Zoos
Decision: Do you need a framework, a managed service, or both?
- Framework: You control more of the application and integration code.
- Managed service: The provider operates substantial parts of the infrastructure.
- Both: Use a framework to compose an application that calls managed models or services.
Checkpoint: Can the tool support testing, version control, deployment, and future maintenance?
D. Infrastructure
Question: Where should the AI system run?
- Managed cloud service: Google Cloud · Amazon · Microsoft
- Specialized model infrastructure: NVIDIA → NGC - NVIDIA GPU Cloud
- On-device or edge computing: TensorFlow Lite → NVIDIA Jetson Nano
- Containerized services: Containers; Docker, Kubernetes & Microservices
- Serverless workflows: Serverless → AWS Lambda & Python
- Enterprise architecture: Enterprise Architecture (EA) → AI Marketplace & Toolkit/Model Interoperability
Next question: What constrains deployment?
- Data cannot leave a controlled environment: Investigate local or approved private deployment.
- Low latency is essential: Evaluate proximity, hardware, model size, and caching.
- Demand varies considerably: Compare scaling and consumption-based pricing.
- High availability is required: Evaluate redundancy, monitoring, recovery, and support.
- Specialized hardware is needed: Examine Processing Units - CPU, GPU, APU, TPU, VPU, FPGA, QPU.
Continue: Compare candidates →
E. Specialized Technology
Question: Which capability is required?
- Document understanding: Document Understanding AI → Retrieval-Augmented Generation (RAG)
- Computer vision: Vision → Getting Started & Project: Object Detection
- Speech interfaces: Speech Recognition → Synthesize Speech
- Robotics and edge AI: Robotics → Embodied AI
- AI-assisted game development: Game Development with Generative AI
- Digital environments: Simulation → Digital Twin
- Security analysis: Cybersecurity → Cybersecurity Companies/Products
Decision: Does the application require a specialized model or can a general-purpose model achieve acceptable performance?
Compare both where practical. Specialized technology may provide better efficiency, control, or domain performance, but the advantage must be demonstrated for the particular task.
F. Compare and Select
Do not begin with a provider preference. Begin with requirements.
For each candidate, assess:
| Criterion | Question |
|---|---|
| Capability | Does it accomplish the required task? |
| Quality | Does it meet evaluation criteria? |
| Economics | What is the total cost of ownership? |
| Integration | Does it work with existing systems? |
| Governance | Can it meet privacy, security, and policy requirements? |
| Operations | Can it be monitored, maintained, and supported? |
| Portability | How difficult would migration to another solution be? |
Decision:
- No candidate meets critical requirements: Reconsider the requirements or solution architecture.
- Several candidates qualify: Test representative workloads and compare results.
- One candidate qualifies: Document the evidence and risks before proceeding.
Next: Build the solution → Validate it
Related Dimensions
- AI Solver — choose the underlying AI approach.
- AI Builder — implement the technology.
- Data & Knowledge Navigator — assess information requirements.
- AI Trust & Evaluation — verify suitability.
- AI Strategy & Adoption — evaluate the investment.