The generative AI landscape is evolving at an unprecedented pace. To build scalable and reliable applications, it is essential to understand the layers that make up the modern AI stack. This is the directory of the key categories.
1. Model Providers
These are the organizations that train and make powerful foundation models (LLMs) available. They deliver raw intelligence and reasoning power via APIs. The choice of model largely determines the capabilities of your final application.
- Anthropic (known for the Claude models)
- OpenAI (known for the GPT series)
- DeepSeek (strong in cost-efficient and open-weight models)
2. Inference & Hosting
Inference providers deliver the computing power (GPUs and cloud infrastructure) needed to actually run AI models. Especially for open-source models, specialized and scalable hosting is crucial for fast response times and affordable processing.
- Together AI
- Hugging Face (Inference Endpoints)
- Groq
3. Vector Databases
To provide models with context via Retrieval-Augmented Generation (RAG), texts are converted into mathematical vectors. Vector databases are specifically built to store this data efficiently and perform lightning-fast semantic searches across millions of documents.
- Pinecone
- Weaviate
- Qdrant
4. Agent Frameworks
Frameworks provide the orchestration layer that allows you to connect LLMs to external tools, memory, and logic. This enables models to not only generate text, but also independently plan and execute multi-step tasks as autonomous 'agents'.
- LangChain
- LlamaIndex
- AutoGen
5. Evaluation Tools
As AI applications become more complex, measuring output quality is indispensable. Evaluation tools help developers monitor hallucinations, test prompts, and ensure consistent performance in production environments.
- LangSmith
- Ragas
- TruEra