Thuta Learning
ရှာဖွေရန်
ExercisesAIbeginner

Context Engineering စိန်ခေါ်မှု အဆင့်မြင့် Drills

စိတ်လျှော့ပါ။ ဒီခန်းကို စာအုပ်လိုမဟုတ်ဘဲ စကားပြောသလိုပဲ၊ နားလည်လွယ်အောင် ရှင်းပါမယ်။

ဒီခန်းပြီးရင် ဘာတတ်သွားမလဲ

  • Context Engineering စိန်ခေါ်မှု အဆင့်မြင့် Drills ကို ကိုယ်တိုင် လေ့ကျင့်ကြည့်မယ်
  • သင်ခဲ့ပြီးသား skill တွေကို practice လုပ်ပြီး ခိုင်မာအောင်လုပ်မယ်
  • အမှားရှာ၊ ပြင်တတ်၊ ကိုယ်တိုင် check လုပ်တတ်မယ်

ခဏလေး ဒီလိုပဲ စဉ်းစားကြည့်

ဒီ lesson က ရှေ့က drill တွေထက် တစ်ဆင့်တက်ပါတယ် — task တစ်ခုချင်းစီမှာ skill တစ်ခုတည်းမဟုတ်ဘဲ Budget the page, Summarize don't dump, Stale notes lie, Missing context and ask back ဆိုတဲ့ chapter တွေက concept ပေါင်းစုံကို ရောပြီး သုံးရပါမယ်။ Token limit ရှိတဲ့ context window ထဲကို information အများကြီးထဲက ဘာကို ဦးစားပေးမလဲ ဆုံးဖြတ်ရတဲ့ trade-off ကို တွေ့ရပါလိမ့်မယ်။ တခြား task တွေက stale (ခေတ်ကျန်နေတဲ့) data ကို ဖမ်းမိနိုင်ဖို့နဲ့ context မလုံလောက်ရင် assume မလုပ်ဘဲ ပြန်မေးဖို့ ကို ဆုံးဖြတ်တဲ့ judgement ကို drill ပေးပါတယ်။ ဒီအဆင့်မှာ 'အကောင်းဆုံး' အဖြေတစ်ခုတည်း မရှိတော့ပါဘူး၊ trade-off ကို ဘယ်လို ချိန်ခွင်လျှာညှိလဲ ဆိုတာချည်းပဲ အရေးကြီးလာပါတယ်။

လေ့ကျင့်ခန်းများ

Task 1 (Budget) - token budget 500 ပဲရှိတဲ့ system အတွက် documents list ထဲက document 4 ခု (word count ပေးထားတယ်) ကို ဘယ်ဟာတွေ ဘယ်လောက် trim ပြီး ထည့်မလဲ ဆုံးဖြတ်ပါ။ Task 2 (Summarize) - chat_history_sample ကို paragraph 1 ခု (sentence 3-4 ကြောင်းအောင်) summarize လုပ်ပြီး key decision တွေချည်းပဲ ကျန်အောင် ရေးပါ။ Task 3 (Stale) - note_with_dates ထဲက note 3 ခုကို date အလိုက် နှိုင်းယှဉ်ပြီး ဘယ်ဟာက stale ဖြစ်နေလဲ ရှာပြီး ဘာလို့ မယုံသင့်လဲ ရှင်းပြပါ။ Task 4 (Missing + ask back) - user_prompt_missing_context အတွက် context ထဲမှာ လိုအပ်တဲ့ key fact (ဘယ် plan ကို upgrade လုပ်မလဲ) ပျောက်နေတာကို ဖော်ထုတ်ပြီး၊ assume မလုပ်ဘဲ ပြန်မေးမယ့် clarifying question တစ်ခု ရေးပါ။

Code နမူနာ

json
{
  "token_budget": 500,
  "documents": [
    {"name": "Product spec v3", "words": 420, "updated": "2026-08-20"},
    {"name": "Product spec v1 (old)", "words": 380, "updated": "2026-05-02"},
    {"name": "Support FAQ", "words": 300, "updated": "2026-07-15"},
    {"name": "Team meeting notes", "words": 250, "updated": "2026-08-22"}
  ],
  "chat_history_sample": [
    "user: dark mode ထည့်ချင်တယ်",
    "assistant: OK, color scheme ဘယ်လိုလုပ်မလဲ",
    "user: system default နဲ့ toggle button ၂ခုလုံးလိုချင်",
    "... (message 8 ခု ကျန်, mostly UI detail ငြင်းခုံမှု)",
    "user: deadline ကို Sept 15 အထိ တိုးလိုက်ပြီ"
  ],
  "note_with_dates": [
    {"note": "Server region: Singapore", "date": "2025-11-01"},
    {"note": "Pricing: $9/month", "date": "2026-08-19"},
    {"note": "Server region: Frankfurt (migrated)", "date": "2026-08-05"}
  ],
  "user_prompt_missing_context": "ကျွန်တော့် account ကို upgrade လုပ်ပေးပါ"
}
You should see
Budget အတွင်းမှာ relevant + up-to-date information ချည်းပဲ ကျန်ပြီး၊ stale note နဲ့ missing fact တွေအတွက် assume မလုပ်ဘဲ clarifying question တစ်ခု ထွက်လာရပါမယ်။

၅ မိနစ် စမ်းကြည့်

အချိန် 5 မိနစ်ပဲ ယူပြီး Task 3 ကို လုပ်ကြည့်ပါ — note_with_dates ထဲက ဘယ် note က stale ဖြစ်နေလဲ ရေးပြီး၊ ဘယ်ဟာကိုသာ context ထဲ ထည့်မလဲ ဆုံးဖြတ်ပါ။

သတိလေးတစ်ချက်

Missing context ကို model ကိုယ်တိုင် guess လုပ်စေတာက အန္တရာယ်အရှိဆုံးပါ — assume လုပ်ပြီး wrong action လုပ်တာထက် ပြန်မေးတာက ပိုကောင်းပါတယ်။

ဒီနေရာမှာ လူအများမှားတတ်တယ်

  • Document အားလုံးကို 'အားလုံးအရေးကြီးတယ်' ဆိုပြီး token budget ကျော်အောင် အကုန်ထည့်လိုက်တာ
  • note ၂ ခုက topic တူနေရင် date ကို မယှဉ်ဘဲ list ရဲ့ ပထမဆုံးတွေ့တဲ့ note ကိုပဲ context ထဲ ထည့်လိုက်တာ

အခု ကိုယ်တိုင် စမ်းကြည့်

အချိန် 5 မိနစ်ပဲ ယူပြီး Task 3 ကို လုပ်ကြည့်ပါ — note_with_dates ထဲက ဘယ် note က stale ဖြစ်နေလဲ ရေးပြီး၊ ဘယ်ဟာကိုသာ context ထဲ ထည့်မလဲ ဆုံးဖြတ်ပါ။

You'll know it worked when: Budget အတွင်းမှာ relevant + up-to-date information ချည်းပဲ ကျန်ပြီး၊ stale note နဲ့ missing fact တွေအတွက် assume မလုပ်ဘဲ clarifying question တစ်ခု ထွက်လာရပါမယ်။

Context Engineering စိန်ခေါ်မှု အဆင့်မြင့် Drills | Thuta Learning