Uploaded October 2026 | Updated November 2026, 3 days ago
You've probably stood in an Aadhaar centre, pressed your fingers on the scanner and stared into the iris camera. What happens after that? Your ten fingerprints and both irises are checked against everyone already enrolled, more than 145 crore Aadhaar numbers, so nobody gets two. In a System Design Interview you get exactly this question: "Design Aadhaar Enrollment System."
Most candidates say "check if the name and date of birth exist, if not save the row and give the next number". Names repeat, people change their details, and a number in sequence is easy to guess.
- Fix 1: capture at the centre. Camps run in villages, so the enrolment client works mostly offline. Details, a face photo, ten fingerprints and both irises. The certified operator signs every packet with their own fingerprint, and it's encrypted on the client before it's saved. You get a slip with a 28-digit enrolment ID. No fingers? Iris only, with a supervisor's sign-off
- Fix 2: an async pipeline, not one API call. The packet uploads to CIDR and moves through checks: tamper, operator, quality, then demographic dedup. UIDAI says up to 90 days. A queue and a workflow, not a request that waits
- Fix 3: 1 : N biometric dedup. Every new enrolment is searched against the entire database: about 600 trillion comparisons a day at the 2014 peak. Three ABIS engines from three different vendors; if one flags a possible duplicate, the others verify it
- Fix 4: a match is a suspicion, not a verdict. Flagged duplicates go to adjudication: clear ones automatically, grey ones to a human team, so no real person is rejected over a false match
- Fix 5: the number comes last. Twelve digits: eleven random, the last a Verhoeff checksum that catches typos. Nothing hidden inside. And by law, core biometrics can't be shared
- The follow-up: one person enrolls at two centres on the same day, and neither packet is in the database yet. Check only against issued numbers and both pass. The fix: add each packet to the search gallery the moment it enters dedup, so the second one hits the first and goes to adjudication
Put together, a strong answer: a signed, encrypted packet, an async pipeline, 1 : N dedup across three engines, a human for the grey cases, and the number last.
Comment and tell me: how long did your Aadhaar take to arrive?
kgptalkie.com
#shorts #SystemDesign #SystemDesignInterview #Aadhaar #UIDAI #Biometrics #Deduplication #BackendInterview #InterviewPrep #SoftwareEngineering #KGPTalkie
You've probably stood in an Aadhaar centre, pressed your fingers on the scanner and stared into the iris camera. What happens after that? Your ten fingerprints and both irises are checked against everyone already enrolled, more than 145 crore Aadhaar numbers, so nobody gets two. In a System Design Interview you get exactly this question: "Design Aadhaar Enrollment System."
Most candidates say "check if the name and date of birth exist, if not save the row and give the next number". Names repeat, people change their details, and a number in sequence is easy to guess.
- Fix 1: capture at the centre. Camps run in villages, so the enrolment client works mostly offline. Details, a face photo, ten fingerprints and both irises. The certified operator signs every packet with their own fingerprint, and it's encrypted on the client before it's saved. You get a slip with a 28-digit enrolment ID. No fingers? Iris only, with a supervisor's sign-off
- Fix 2: an async pipeline, not one API call. The packet uploads to CIDR and moves through checks: tamper, operator, quality, then demographic dedup. UIDAI says up to 90 days. A queue and a workflow, not a request that waits
- Fix 3: 1 : N biometric dedup. Every new enrolment is searched against the entire database: about 600 trillion comparisons a day at the 2014 peak. Three ABIS engines from three different vendors; if one flags a possible duplicate, the others verify it
- Fix 4: a match is a suspicion, not a verdict. Flagged duplicates go to adjudication: clear ones automatically, grey ones to a human team, so no real person is rejected over a false match
- Fix 5: the number comes last. Twelve digits: eleven random, the last a Verhoeff checksum that catches typos. Nothing hidden inside. And by law, core biometrics can't be shared
- The follow-up: one person enrolls at two centres on the same day, and neither packet is in the database yet. Check only against issued numbers and both pass. The fix: add each packet to the search gallery the moment it enters dedup, so the second one hits the first and goes to adjudication
Put together, a strong answer: a signed, encrypted packet, an async pipeline, 1 : N dedup across three engines, a human for the grey cases, and the number last.
Comment and tell me: how long did your Aadhaar take to arrive?
kgptalkie.com
#shorts #SystemDesign #SystemDesignInterview #Aadhaar #UIDAI #Biometrics #Deduplication #BackendInterview #InterviewPrep #SoftwareEngineering #KGPTalkie










