DigiLocker AI Agent : My Old Suggestions Find Their Moment
· @Hemen Parekh
==============================================
Context
Govt close to picking private AI integrator for DigiLocker / Economic Times / Oct 2026
Gist (paraphrased from several reports):
- MeitY has shortlisted CoRover and Kyndryl Solutions to build a reusable agentic AI platform for government services, with DigiLocker as the first use case (AIM, 23 Sept 2026).
- The agent follows an "ask, do and escalate" design: answer citizen queries, carry out transactions, and hand difficult cases to a human. Build time is four months, then 12 months of operations.
- The architecture is to support DigiLocker / Parichay authentication, consent management and audit trails, in multiple languages (NewsBytes).
- NeGD's earlier AI tender lists conversational and voice-based form filling, face matching with liveness detection, and agents that outside developers can embed in their own apps (NeGD RFE).
- Separately, a draft Request for Empanelment (21 Sept 2026) seeks agencies to bring banks, insurers, educational institutions and hospitals onto DigiLocker as document requesters. Comments are due 10 October 2026 (MediaNama).
My Take
In July 2023, in Digilocker : the buck stops here, I listed 12 e-mails I had sent to Cabinet Ministers on DigiLocker and a single citizen identity, going back to May 2015.
This is a follow-on. I am leaving aside every part of MeitY's current plan on which I never commented (escalation to humans, audit trails, vendor lock-in, global credential verification, the choice of AI tools). Below are only those seven aspects where, years ago, I had something specific to suggest.
Seven aspects, then and now
1. An agent that "does", so the citizen fills no form
Now: the agent will carry out transactions, and NeGD lists voice-based form filling.
Then: in I am One, I will become Many (6 Jan 2020) I laid down one condition: merging a citizen's cards must NOT require him to fill up any form, online or offline.
Suggestion today: make "zero forms" the measurable target. Every service the agent handles should be judged by how many fields the citizen still types.
2. Creating a new card, not just storing old ones
Now: "do" means transactions, not only document fetching.
Then: same 2020 blog, I asked: why not let a citizen use the app to create a brand-new card, e.g. a Voter Card on turning 18?
Suggestion today: after the DigiLocker pilot, the first "do" use cases should be first-time issuance: Voter ID at 18, PAN, Ayushman card, using documents already in the locker as proof.
3. Talk to it in your own language
Now: conversational and voice AI, multilingual, for UMANG and DigiLocker.
Then: in Digital Locker Project (23 May 2015) I urged Shri Ravi Shankar Prasad to drop the crowd-sourced typing scheme and build a "DigiDoc" app where a citizen simply reads his document aloud, in any official Indian language, with self-learning speech-to-text. In March 2022, welcoming the WhatsApp DigiLocker chatbot, I urged Shri Amit Ranjan to keep evolving it.
Suggestion today: voice must work for the whole journey, not just queries. And let citizens' corrections train the model, as I proposed in 2015.
4. Your face as the key
Now: face matching with liveness detection, and AI-based eKYC.
Then: my YUP (Your Unique Profile) app: open it, enter mobile number, take a selfie, and the system matches the face to retrieve all your cards (2020, repeated in DigiLocker 2.0, Feb 2023).
Suggestion today: offer selfie + liveness as a full alternative to OTP, especially for elderly users whose mobile numbers keep changing.
5. Edit once, updated everywhere
Now: DigiLocker 2.0 already promises that a change flows to all linked documents; Aadhaar address changes now update the driving licence.
Then: YUP let the citizen edit his personal data online and click SUBMIT, once. I welcomed the first step in Morphing of DigiLocker : Caterpillar to Butterfly (17 Mar 2022).
Suggestion today: give the agent this as a standard skill: "I have moved house" should update every issuer in one conversation, with consent per issuer.
6. Show it to any party
Now: the RFE brings banks, insurers, colleges and hospitals on board as requesters.
Then: YUP's last step: the citizen can show his profile to any party, any time, instead of photocopies.
Suggestion today: each new requester should accept DigiLocker documents as final proof, and stop asking for physical copies "as well".
7. One identity for businesses too
Now: Entity Locker and enterprise DigiLockers for MSMEs and trusts.
Then: An "Aadhar" for Corporate ? (12 May 2016) listed 16 separate numbers a company needs, and asked for one simple identity.
Suggestion today: the same agent framework should serve MSMEs: one conversation to file, renew or update across GST, EPFO, ESIC and Udyam.
Voluntary, always
The plan stresses consent management. My 2020 framework said the same: YUP is absolutely voluntary, and a citizen may keep using physical cards if he wishes.
At a glance
Current thinking (2026) | What I suggested | First blogged |
|---|---|---|
Agent carries out transactions; voice form filling | No form to fill, online or offline | Jan 2020 |
"Do" beyond fetching documents | Create a new card, e.g. Voter ID at 18 | Jan 2020 |
Multilingual conversational / voice AI | DigiDoc: read documents aloud in any Indian language | May 2015 |
Face match with liveness; AI eKYC | YUP selfie retrieves all your cards | Jan 2020 |
Change once, all linked documents update | Edit personal data once, click SUBMIT | Jan 2020 |
Banks, insurers, colleges, hospitals as requesters | Show YUP to any party | Jan 2020 |
Entity Locker / enterprise lockers | One identity for companies instead of 16 numbers | May 2016 |
Consent management | YUP absolutely voluntary | Jan 2020 |
Where I would refine my own idea
In 2020 I proposed that all card-issuing agencies upload their databases daily to one SUIIC server, to be merged into a single profile. MeitY has chosen differently: each issuer keeps its data, DigiLocker fetches it with the citizen's consent, and the AI agent works on top.
I accept that this federated route is wiser after the Puttaswamy judgment and the DPDP Act. It delivers most of what YUP promised, without one giant database to guard. My suggestions above are written for that architecture.
A request to NeGD
The draft RFE on private-sector onboarding is open for comments until 10 October 2026. I will be sending these seven suggestions to NeGD. Readers who agree may do the same, with their own improvements.
With regards,
Hemen Parekh
www.hemenparekh.ai / 05 Oct 2026
Related readings: Digilocker : the buck stops here (Jul 2023) · DigiLocker 2.0 = I am One, I will become Many (Feb 2023) · Meri Aawaz hi Meri Pehchaan Hai (Jul 2022) · DigiLocker = Single Unified Identity for Indian Citizen ? (Mar 2022) · Digital Locker Project (May 2015)
No comments:
Post a Comment