Real systems and frameworks worth studying once the capstone is done.
| System | What to look for |
|---|---|
| Anthropic — Building Effective Agents | Orchestrator–subagent patterns, task decomposition, handoffs |
| LinkedIn RAG at Scale | Hybrid search, retrieval monitoring in production |
| Uber LLM Gateway | Model routing, cost attribution, fallback across providers |
| Replit AI Agent | Long-horizon agent with persistent state and sandboxed tool execution |
| Lilian Weng — LLM Powered Autonomous Agents | The most complete survey of agent patterns: planning, memory, tool use |
| OWASP Top 10 for LLM Applications | Security risks specific to LLM systems — prompt injection, data leakage, supply chain |
Study as reference implementations, not mandatory dependencies.
| Framework | What to study |
|---|---|
| LangGraph | Stateful graph-based orchestration |
| LangChain | RAG, tool calling, and chain abstractions — large ecosystem, use selectively |
| LlamaIndex | Data ingestion, indexing, and retrieval — strongest RAG-specific framework |
| AutoGen | Multi-agent conversations: planner, executor, and critic |
| CrewAI | Role-based agent teams with task delegation |
| Anthropic Agent Patterns | Orchestrator–worker, parallelisation, routing — no framework lock-in |
| Vercel AI SDK | Streaming, tool calling, and UI integration for TypeScript/Next.js |
| Haystack | Modular RAG and agent pipelines — good for custom retrieval workflows |