Learn AI — from tokens to agents to evals.
Migrated from flexappdev.com/bo2 — tokens and embeddings through agents, RAG, and evals.
BPE, SentencePiece — why your bill isn't measured in words.
Vector representations — semantic search, clustering, RAG retrieval.
Self-attention, positional encoding, decoder-only — the architecture eating the world.
1M tokens with Claude 4.7 — how to actually use that real estate.
5-min TTL on the Anthropic API — cache hits cut latency and cost.
Build custom agents on Anthropic's managed infrastructure.
Anthropic's pre-built agent harness on managed infrastructure.
Model Context Protocol — standard way to plug tools into agents.
Reusable invocable workflows — the building blocks of super-agents.
Spawn isolated loops that return summaries to a master agent.
Deterministic scripts that run outside the loop on tool events.
Orchestrators, specialists, trust boundaries — scaling beyond one loop.
Retrieval-augmented generation — embeddings + reranking + careful chunking.
Function calling done right — schemas, validation, parallel tool calls.
Measure skill performance with variance analysis, not vibes.
Persistent agent memory — cognee-style stores, structured recall.
ReAct, plan-and-execute, reflection — the design vocabulary for agents.
JSON Schema-constrained responses — reliable parsing from unreliable models.
Short-term, long-term, semantic — persistent recall across sessions.
Prompt injection, jailbreaks, output validation — the agent attack surface.
Nano Banana (Gemini CLI), diffusion models — agents shipping pictures.
Programmatic Remotion + AI cuts — the agent-driven content pipeline.
TTS / STT / voice agents — Whisper, ElevenLabs, real-time pipelines.