How Aadhaar Identifies a Billion People — System Breakdown Ep 7
In this episode of System Breakdown, we delve into the intricacies of the Aadhaar system, a digital public infrastructure that identifies over a billion people in India. The Aadhaar system design is a marvel of backend engineering, utilizing biometric identification and deduplication to ensure unique identities. The process of Aadhaar authentication involves a complex interplay of registered devices, data retention, and high availability, making it a fascinating example of distributed systems in action. We explore how senior engineers think about system design, and the trade-offs between identification and verification in the context of biometric data. The Aadhaar system also features innovative solutions such as offline verification and QR code-based identification, making it a robust and reliable digital infrastructure. By examining the Aadhaar system design and its various components, including software architecture and system breakdown, we can gain a deeper understanding of how this massive system works, and what lessons it holds for the design of similar systems in the future. The episode is part of a series that aims to explain complex system design concepts in an approachable way, using the example of Aadhaar to illustrate key ideas and principles.
Chapters
0:00 Two questions, two machines
0:21 Verification vs identification
1:23 The scale
2:10 What counts as the core
2:57 The whole architecture: three paths
4:13 Capture and the encrypted packet
5:16 Validate first, and the forever archive
6:29 Population-scale 1:N search
8:14 Three biometric engines
9:48 Human review, and the under-five exception
11:10 The identifier
12:02 Who is allowed to ask
13:16 The fast path: 1:1 authentication
14:51 Trust at the sensor
16:17 Verify by signature, no server
17:22 Four retention clocks
19:15 The join key: Virtual ID and UID Token
20:28 The hard problem
21:55 Thresholds and failure paths
23:49 Serving capacity
25:55 What we know, and what we don't
27:26 Five system-design moves
Sources
· UIDAI, Aadhaar Generation page: generation stages, the "forever" packet archive, demographic de-duplication (including under-fives), three ABIS vendors, anonymised biometrics, manual adjudication
· UIDAI Annual Report 2024–25, §7.3.2: HD → UHD data centres, racks 260 → 80, open-source private cloud, 600 biometric blade servers (200 each NEC / Idemia / TCS), ~30 PB migration, MySQL 8 InnoDB Cluster
· UIDAI, About Authentication: online authentication and e-KYC active-active across Hebbal and Manesar
· Aadhaar (Authentication and Offline Verification) Regulations, 2021: 1:1 authentication, AUA/ASA roles, retention periods for UIDAI, requesting entities and ASAs
· Aadhaar Authentication API 2.5 and Registered Devices specification 2.0
· Aadhaar Act 2016, Section 32(3); Supreme Court, Puttaswamy (2018)
· UIDAI enrolment dashboard (May 2026): 144.66 crore Aadhaar generated
· UIDAI Chairman at the Global Fintech Festival (Sep 2026): ~10 crore authentications a day, 25–30 crore/day preparation target, 600+ entities
· PIB (Jul 2010): original multi-ABIS design; PIB (Sep 2026): Face Authentication SDK
· Department of Food & Public Distribution FAQs: alternatives when biometric authentication fails
Next episode: WhatsApp. More than three billion people, and personal chats it can't read.
⚠️ Educational system-design teardown. Figures are point-in-time and sourced as above. The 25–30 crore/day figure is a stated preparation target, not measured throughput.