Lesson · Hinglish · ⏱ 8 min padhne mein
🔐 Enterprise AI: integration, security aur compliance
EnglishHinglish
🏠 Asli zindagi ki misaal
Hospital visiting specialist ko andar aane deta hai, lekin aise visitor badge ke saath jo sirf kuch darwaaze kholta hai, har entry log hoti hai, aur koi patient file building se bahar nahi jaati. Company mein AI laana bhi aise hi kaam karta hai: woh legacy systems se controlled darwaazon ke through judta hai, har user ko sirf wahi dikhata hai jo use dekhne ki permission hai, aur audit trail chhodta hai.
💡 Aasaan shabdon mein
Enterprise AI projects model se nahi, integration aur bharose se safal ya fail hote hain. Existing systems (CRMs, ERPs, document stores) se unki APIs ke through, sahi authentication (SSO, OAuth, service accounts) ke saath judo. RAG ko permission-aware banao: sirf woh documents retrieve karo jo poochne wale user ko dekhne ki permission hai. Personal data (PII) ko model tak pahunchne se pehle mask karke bachao, aur pata rakho ki data kahan store aur process hota hai (data residency). Audits ke liye har prompt, tool call aur jawaab log karo. Client ki security team ke saath shuru mein hi requirements ko GDPR, India ke DPDP Act, HIPAA ya SOC 2 jaise frameworks se map karo.
🪜 Step by step
- Data ka map banao: AI system kaunsa personal ya confidential data chhoota hai, aur woh kahan jaata hai?
- Rules poore karne wale deployment aur vendors chuno: data residency, aapke data pe training nahi, signed agreements.
- Company ka identity system (SSO) aur permissions use karo, taaki AI users ko kabhi usse zyaada na dikhaye jitna woh pehle dekh sakte the.
- Jahan zaroori na ho wahan personal data mask ya hatao, aur secrets vault mein rakho.
- AI ke actions aur decisions ke audit logs rakho, aur reviews ke liye risks aur controls document karo.
🔑 Zaroori shabd
- PII
- Personally identifiable information: naam, phone number, Aadhaar, address.
- Permission-aware retrieval
- Search jo har user ke access rights ka dhyaan rakhe.
- Audit log
- Kisne kab kya kiya, iska aisa record jise badla na ja sake.
👀 Example dekho
HR assistant asks: "What is Priya's salary?"
1. Identify the user via SSO → role = 'team lead'
2. Retrieval filter: only documents this role can read → salary files excluded
3. Answer: "I can't share individual salary details."
4. Audit log: user, question, documents checked, answer, timestampSecurity system lagaata hai. Model ke mana karne ki ummeed pe nahi chhodi jaati.
⚠️ Aam galtiyaan jinse bachna hai
- Security review aakhir tak chhodna. Security aur legal teams ko jaldi shaamil karo; mahine bachte hain.
- Prototyping ke dauraan customer data unapproved AI tools ko bhejna.
✅ Jaldi se check karo
Do employees same RAG bot se confidential plans ke baare mein poochte hain. Kya alag hona chahiye?
- Sirf font
- Har insaan ki permissions ke hisaab se retrieve hone wale documents
- Kuch nahi, dono ko same jawaab
Jawaab dekho
B. Har insaan ki permissions ke hisaab se retrieve hone wale documents: Permission-aware retrieval yeh pakka karta hai ki har user sirf wahi content dekhe jiski use permission hai.
🙋 Seekhne wale yeh poochhte hain
India mein kaunse kaanoon matter karte hain?
Digital Personal Data Protection Act (DPDP) personal data pe laagu hota hai, saath mein sector rules (jaise banking ke liye RBI). Global customers ke saath GDPR aur doosre bhi.
Kya confidential data ke saath cloud LLMs use kar sakte hain?
Aksar haan, enterprise agreements, sahi region, aapke data pe training nahi aur sahi access controls ke saath. Kuch cases mein abhi bhi self hosted models chahiye.
📌 Yaad rakho
Access enforce karo, PII mask karo, sab log karo, aur security team ko jaldi shaamil karo.
📖 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