# 6. Log + eval log_to_bigquery(query, response, contexts, user_session) return response
Artificially expanding datasets to make models more robust. ai.102
| Category | Tools | |----------|-------| | Orchestration | LangChain, LlamaIndex, Haystack, DSPy | | Evaluation | DeepEval, RAGAS, Phoenix Arize, LangSmith | | Structured output | Instructor (Python), Outlines, Guidance | | RAG evaluation | Ragas, TruLens | | Guardrails | Guardrails AI, NeMo, Llama Guard, NeMo | | Observability | Weights & Biases, Langfuse, Honeycomb for LLMs | Fix: Users click thumbs up even when answer
The transition to ai.102 architecture is already reshaping industries. Because of its efficiency and enhanced reasoning capabilities, it is enabling use cases that were theoretically possible but practically unfeasible just two years ago. # 6. Log + eval log_to_bigquery(query
AI.102 is where you learn that an LLM is a stochastic system—and you need deterministic boundaries.
Symptom: Thumbs up/down from users trusted blindly. Fix: Users click thumbs up even when answer is wrong (just convenient). Use periodic blind reevaluation.