AI & Agentic Systems
LLM serving, retrieval, vector search, agent memory, tool harnesses, orchestration, evals, and AI security.
-
LLM Serving & Inference Infrastructure
Prefill vs decode, the KV cache, continuous batching, streaming, GPU capacity and cold starts, and model routing.
-
Token Economics, Caching & Quotas
Per-request cost, prompt caching, semantic caching, context budgets, per-user quotas, and cheaper-model fallback.
-
RAG Architecture & Retrieval
Chunking, embeddings, hybrid retrieval, reranking, context assembly, citations, and when not to use RAG.
-
Vector Databases & ANN Indexes
Exact vs approximate search, HNSW and IVF-PQ, recall vs latency, filtered ANN, sharding, and pgvector.
-
Agent Memory Systems
Working, episodic, and semantic memory, write policies, deduplication, decay, retrieval, and privacy.
-
Agent Harness, Tools & Sandboxing
Tool schemas and dispatch, step limits, sandboxed execution, filesystem and network policy, MCP servers, and permissions.
-
Multi-Agent Orchestration & Workflows
Fan-out and fan-in, orchestrator and workers, context isolation, deterministic vs model-driven routing, and durable execution.
-
Evals, Guardrails & LLM Observability
Offline evals, LLM-as-judge, golden datasets, A/B and shadow traffic, tracing, and drift after model upgrades.
-
Prompt Injection & AI Security
Direct and indirect prompt injection, the lethal trifecta, untrusted output, exfiltration paths, and least-privilege tools.
Up next
49 of 49 questions here are still unsolved.