Agents को evaluate करना
किसी agent के लिए eval suite बनाओगे: realistic tasks, outcome और trajectory के graders, consistency नापने के लिए बार-बार runs, और ऐसा tracking जो असली progress दिखाए।
Pro 1 · Quality engineerabout 4 hours7 sections
सीधा जवाब
Agents को realistic tasks पर evaluate करो: final outcome और उठाए गए steps दोनों grade करो, consistency नापने के लिए हर task कई बार चलाओ, और cost और time भी track करो।
ये क्यों ज़रूरी है
Agents एक अकेली model call से ज़्यादा तरीकों से fail होते हैं: गलत tools, बेकार steps, unsafe actions, और किस्मत से सही answer। Agent-specific evals के बिना हर prompt change एक जुआ है। Agent का काम देने वाली teams पूछती हैं कि आप एक को कैसे evaluate करोगे।
Sign up free to read the whole lesson
The rest of this lesson: 7 parts, video lessons, the best course, practice and a quick check. It is free, takes a few seconds and needs no card.
Accounts are for learners aged 18 and over. Under 18? The quick answer and the first part of every lesson stay open: browse lessons by topic.
Lessons on Agentic AI Engineer0 of 27 done
- 1Call an LLM API from code
- 2Get structured JSON output
- 3Tool calling
- 4The agent loop
- 5Designing tools for agents
- 6Agent memory
- 7Planning and reflection
- 8Context engineering
- 9LangGraph: agents as graphs
- 10Agent SDKs and sub-agents
- 11Multi-agent systems
- 12MCP servers and clients
- 13Browser and computer-use agents
- 14Logs and traces for agents
- 15How agents fail
- 16Test sets and evals
- 17Defend against prompt injection
- 18Human approval steps
- 19Guardrails
- 20Permissions and sandboxing
- 21Loop limits, budgets and kill switches
- 22Evaluating agents
- 23Advanced harness engineering
- 24Harnesses for long-running agents
- 25Deploy and monitor an LLM app
- 26Cost control
- 27Accountability for agent actions
Related
On the roadmap: Agentic AI Engineer
EvaluationPro 1: Quality engineerVideo lessonsTrajectoryEvalsLLM as judgeGolden set