A landscape survey of the tools, platforms, and frameworks that make up the modern AI engineering stack.
The Full Stack
graph TD
A[LLM Providers · OpenAI · Anthropic · Google · Meta] --> B[Orchestration Frameworks]
B --> C[LangGraph · LangChain · CrewAI · AutoGen · Spring AI]
C --> D[Tool & Integration Layer]
D --> E[MCP Servers · GitHub · JIRA · Playwright · Vector DBs]
D --> F[Observability · LangSmith · OpenTelemetry · Langfuse]
D --> G[Guardrails · LlamaGuard · Guardrails AI · custom validators]
LLM Provider Comparison
| Provider |
Best Models |
Context |
Function Calling |
Self-hosted |
| OpenAI |
GPT-4o, o3-mini |
128K |
Yes, structured outputs |
No |
| Anthropic |
Claude 3.5 Sonnet, Claude 4 |
200K |
Yes, tool use |
No |
| Google |
Gemini 1.5 Pro |
1M |
Yes |
Via Vertex AI |
| Meta |
LLaMA 3.3 70B |
128K |
Yes (via llama.cpp) |
Yes |
| Mistral |
Mistral Large 2 |
128K |
Yes |
Yes |
| Cohere |
Command R+ |
128K |
Yes, RAG-optimised |
No |
Orchestration Frameworks
| Framework |
Language |
Style |
Best For |
| LangGraph |
Python |
Explicit graph |
Production agents, complex flow |
| LangChain |
Python |
Chain/pipe |
RAG pipelines, simple agents |
| CrewAI |
Python |
Role-based agents |
Multi-agent prototyping |
| AutoGen |
Python |
Conversational multi-agent |
Research, prototyping |
| Semantic Kernel |
C# / Python |
Plugin-based |
.NET ecosystem integration |
| Spring AI |
Java |
Spring-native |
Java dev teams, Spring Boot |
| Haystack |
Python |
Pipeline-focused |
Enterprise RAG, search |
Vector Databases
| Database |
Hosting |
Max Scale |
Spring / Java SDK |
| Pinecone |
Managed |
Billions |
REST API |
| Weaviate |
Self/cloud |
Hundreds of millions |
Java client |
| Qdrant |
Self/cloud |
Hundreds of millions |
REST/gRPC |
| ChromaDB |
Self |
Millions (dev only) |
REST API |
| pgvector |
Postgres |
Tens of millions |
JDBC (Spring Data) |
| OpenSearch/kNN |
Self/cloud |
Billions |
Spring Data OpenSearch |
→ Deep Dive: Coding Agents
→ Deep Dive: JIRA Integration
→ Deep Dive: CI/CD Integration
| Tool |
Purpose |
| LangSmith |
Tracing, evaluation, prompt management for LangChain/LangGraph |
| Langfuse |
Open-source LLM tracing, works with any framework |
| Phoenix (Arize) |
LLM observability, prompt evaluation |
| RAGAS |
RAG pipeline evaluation (faithfulness, relevance, recall) |
| Braintrust |
Evaluation framework with human feedback loop |
| Helicone |
LLM proxy with logging, cost tracking |
Guardrails & Safety
| Tool |
What It Checks |
| Guardrails AI |
Output schema validation, content filters, custom validators |
| LlamaGuard |
LLM-based safety classifier for inputs and outputs |
| NeMo Guardrails |
Conversational guardrails for topic and behaviour control |
| PromptArmor |
Prompt injection detection |
| Custom validators |
Post-process output, check diff size, validate syntax before applying |
| Tool |
What It Does |
MCP Support |
| GitHub Copilot |
In-editor code completion and chat |
Yes (Agent mode) |
| Cursor |
AI-first IDE, context-aware editing |
Yes (native MCP client) |
| Aider |
CLI coding agent using git diff approach |
Limited |
| Continue.dev |
VS Code / JetBrains extension, custom models |
MCP via config |
| Codeium |
Free Copilot alternative, fast completions |
No |