Lesson · Hinglish · ⏱ 7 min padhne mein
📋 Eval plan ke saath AI product spec (PRD) likhna
EnglishHinglish
🏠 Asli zindagi ki misaal
Kitchen renovate karne ke liye contractor rakhte waqt, acche brief mein hota hai: kya chahiye, budget, zaroori cheezein, kya nahi karna, aur kaam kaise check karoge. AI PRD AI feature ke liye wahi brief hai, ek extra hisse ke saath jo normal software ko nahi chahiye: 'zyaadatar sahi' ho sakne wale jawaabon ko kaise judge karoge.
💡 Aasaan shabdon mein
Acche AI PRD mein hota hai: user aur problem, success metric, scope aur non-goals, aur user experience, jismein yeh bhi ki AI unsure ya galat ho to kya hoga. AI-specific sections bhi jodo: model ko kaunsa data aur context chahiye, approach (prompting, RAG, fine-tuning), eval plan (eval set, quality bar, kaise naapenge), risks (hallucination, privacy, bias, prompt injection) aur unke mitigations, cost aur latency budgets, human oversight, aur launch criteria. Banaane se pehle eval plan likhne se sab 'accha' ka matlab ek jaisa maan lete hain.
🪜 Step by step
- Problem aur users: dard kise hai, kitna bada hai, aur aaj ka baseline.
- Solution aur scope: AI kya karta hai, kya nahi, aur kahan insaan loop mein rehte hain.
- Quality bar: eval set, kaise score hota hai, aur launch thresholds (zero tolerance cases samet).
- Risks aur unke upaay: galat jawaab, privacy, misuse, cost; aur har ek ka fallback.
- Metrics aur rollout: layered success metrics, guardrails, phased launch plan aur review date.
🔑 Zaroori shabd
- Non-goals
- Cheezein jo feature jaan-boojhkar nahi karega.
- Quality bar
- Launch ke liye zaroori minimum eval score.
- Launch criteria
- Live jaane se pehle poori hone wali shartein.
👀 Example dekho
AI PRD: Policy assistant for branch staff
Problem: staff spend ~25 min/day searching policy PDFs
Metric: time-to-answer < 1 min; 90% answers correct (eval set of 100 real questions)
Approach: model API + RAG over policy docs, citations required
Non-goals: no customer-facing use in v1; no advice outside the policies
Risks: wrong answer → always cite + 'confirm with manager' for loans > ₹10L
Launch: ≥ 90% on evals, < 3 s p95 latency, legal sign-offEval set aur quality bar code likhne se pehle hi tay ho jaate hain.
⚠️ Aam galtiyaan jinse bachna hai
- Bina naapne layak definition ke "AI accurate hona chahiye" likhna.
- AI fail ho to kya hoga yeh chhod dena. Aksar spec ka sabse zaroori hissa yahi hota hai.
✅ Jaldi se check karo
Aisa kaunsa key section hai jo AI PRD mein chahiye lekin normal PRD mein aksar nahi hota?
- Competitors ki list
- AI outputs ke liye eval plan aur quality bar
- Colour palette
Jawaab dekho
B. AI outputs ke liye eval plan aur quality bar: AI outputs badalte rehte hain, isliye pehle se define karna padta hai ki quality kaise naapoge aur kitna accha kaafi hai.
🙋 Seekhne wale yeh poochhte hain
AI PRD normal PRD se kaise alag hai?
Isme eval plan aur quality thresholds, saaf failure handling, data ki zaroorat, per use cost aur model change management judte hain.
Eval set kaun likhta hai?
Aam taur pe PM, domain experts aur engineers milkar. Domain experts tay karte hain sahi jawaab kya hai.
📌 Yaad rakho
Problem, metric, scope, failure ke liye UX, eval plan, risks, launch bar.
📖 Is lesson ke shabd (AI dictionary)
Poora course, aapke roadmap ke hisaab se
Progress tracking, quick checks aur step-by-step order ke saath. Yeh lesson in courses ka hissa hai:
Course mein seekho