TypeScript-Only AI Stack
“Primary SDK is Python-first; TypeScript port lags significantly behind”
LLM Production Infra Stack
“Abstractions add complexity when LiteLLM already handles the provider normalization layer”
Multi-Modal RAG Stack
“LlamaIndex's multimodal parsing is significantly more mature for mixed-format documents”
AI Red-Team / Security Stack
“Orchestration framework — not designed for adversarial prompt testing”
Fine-Tuning Pipeline
“Inference orchestration layer — irrelevant during the training phase”
Document Intelligence Stack
“General orchestration; LlamaIndex's document parsing is purpose-built for this extraction pattern”
Memory-Augmented Agent Stack
“LangGraph provides tighter per-step state control for multi-turn agents that need memory at each decision point”
AI Guardrails Stack
“Orchestration framework; guardrails sit below the application code at the infrastructure layer”
Graph RAG Stack
“General orchestration layer; LightRAG's dual-level retrieval is self-contained and doesn't require a separate chain orchestrator”
Vibe Coding Stack
“Python-first framework; Lovable's generated stack is TypeScript/React — adding Python introduces a language boundary before the product is proven”