Platforms
Platform guides for building AI solutions. Each covers the service landscape, architecture patterns, cost models, and implementation detail.
| Platform | Coverage | Status |
|---|---|---|
| Microsoft | Foundry, OpenAI, AI Search, Agents, Copilot Studio, Fabric, Security, GenAI Gateway | Active |
| Databricks | Mosaic AI, MLflow, Vector Search, Unity Catalog, Genie | Active |
| Snowflake | Cortex Functions, Cortex Analyst, Cortex Search | Active |
| AWS | Bedrock, SageMaker, Textract, Kendra | Active |
| GCP | Vertex AI, Gemini, ADK, Document AI | Active |
- Platform Comparison - side-by-side assessment across capability domains and cost.
- Target Architectures - reference architectures for agentic AI per cloud.
Model and Service Catalogue
The platform pages above cover the major clouds in depth. In practice, delivery teams draw on a wider set of tools across six categories - a model layer, the platforms that serve those models, and the surrounding frameworks for orchestration, retrieval, modality handling, and safety.
| Category | Tools |
|---|---|
| Models | OpenAI, Anthropic, Meta, Cohere, Qwen |
| Model Platforms | Azure Foundry, Amazon Bedrock, Vertex AI, Snowflake Cortex AI, Databricks |
| Agentic Framework & Orchestration | Microsoft Agent Framework, LangChain, LangGraph, Amazon Bedrock Agents, Google ADK, Agent 365, n8n |
| RAG & Vector Store | Azure AI Search, Pinecone, Chroma, ElasticSearch |
| Modality-Specific (Speech & OCR) | Whisper, Amazon Textract, Tesseract OCR, Paddle OCR |
| Observability, Evaluation & Safety | LangSmith, Datadog, Azure AI Content Safety, Guardrails AI, AWS Guardrails |
Illustrative selection of tools deployed on client engagements, not an exhaustive list or an approved-vendor register. Accurate as at August 2026.
Two things worth reading off this table. The model layer and the model platform layer are separate choices - Anthropic and Meta models are reachable through Bedrock, Vertex AI, and Foundry alike, so picking a platform does not lock in a model vendor. And the last two rows are the ones most often underestimated in scoping: modality handling and the observability/safety stack are frequently treated as afterthoughts, then turn into the bulk of the hardening work before a pilot can go to production.