Who we are

SWAN Digital Foundation is a not for profit company registered under Section 8 of the Companies Act, 2013, with its registered office in New Delhi. It was incorporated on 27 November 2007 as Eko Aspire Foundation and was renamed SWAN Digital Foundation with effect from 1 January 2025.

For most of that history the Foundation worked on the economic wellbeing of the urban and rural poor, largely through self help groups. That object remains part of our constitution. What has changed is the instrument. The problems we work on are now problems that software, artificial intelligence and Indian language voice technology can address at a cost per person that traditional programme delivery cannot reach, and our Memorandum is being revised to say so plainly.

What we are working on

India solved the plumbing of financial inclusion. Bank accounts were opened at a scale no country has matched, payment rails reach every district, and identity verification takes seconds. What was not solved is the human layer. A bank account is opened by a person, in a village, sitting at a small counter with a fingerprint scanner and a phone. That person is a Business Correspondent agent running a Customer Service Point, and there are more than a million of them.

Increasingly, and by deliberate policy, that person is a woman. State rural livelihoods missions have placed tens of thousands of women in these roles as BC Sakhis. It is a good policy. It puts income in women's hands, and it puts a woman at the counter, which changes who is willing to walk up to it.

The policy stops short of what it needs. A woman is selected, trained for a few days, given a device and a bank identifier, and then expected to build a business. Nobody teaches her how to bring customers in. Nobody helps her when a transaction fails and a customer is standing in front of her. She cannot travel to meet a supervisor, so she does not. She cannot spare an evening for a training session, so she does not attend. Her business plateaus, her commission stays low, and within a year or two a large share of women agents stop working.

Four things stand in her way, and they compound. She has no discretionary hour in the day. She meets resistance to spending time on work seen as outside the home. She has been taught to doubt her own competence with technology and with money. And there is no structured business support of any kind sitting between her and the bank.

We build against all four, in software, on the phone she already owns, in the language she speaks, at a cost per operator low enough that a state mission or a bank can afford to run it for everybody.

Our mission

To make digital technology genuinely useful to women at the last mile, so that they earn more, decide more and are counted, and to place every tool we build in the public domain so that others can do the same at greater scale than we ever will.

What we are not

We are not a network operator. We do not recruit agents, we do not hold a Business Correspondent contract with a bank and we do not take a share of transaction revenue. We deploy onto cadres that other institutions have already built.

We are not a software vendor. We do not sell licences, we do not charge for the platform and we do not hold anything back in a paid tier. The core is being released under the GNU Affero General Public License version 3, which means anyone can run it, and anyone who improves it and offers it as a service must publish their improvements.

We are not a training provider. Classroom training is not the constraint. Being alone at the counter when something goes wrong is the constraint.

We are not the commercial arm of anybody. Our relationship with Eko Bharat Ventures Private Limited is set out in full on a page of its own, including what Eko does for us, what we pay, the cap on that payment and who approves it.

Our history in brief

This is our principal programme. It puts a coach, a marketing team and a customer relationship system into the hands of a woman running a banking outlet, delivered entirely through the phone she already owns, and priced so that a state mission or a bank can extend it to an entire cadre rather than to a pilot cohort.

The platform is called TARA, Trusted Agents for Rising Aspirations. It has two parts, and they solve two different problems. The first brings new customers to her counter. The second keeps the customers she already has.

Part one, the coach that helps her be seen

A woman agent cannot market her business the way a man in the same role does. She cannot spend the evening at the tea stall telling people what services she offers, and she often cannot travel to the neighbouring hamlet at all. Her business is limited to the people who happen to walk past her door.

Our first tool removes that limit. It helps her produce short, local language content about her services, in her own voice and under her own name, and distribute it through the social and messaging platforms her community already uses. She records a few seconds of audio or answers a few prompts, and the system produces something she can send to a group, post to a status or share with a self help group federation. She does not learn video editing. She does not learn marketing. She speaks, and the work is done for her.

The point is not the content. The point is that a woman who cannot leave her counter acquires a reach that previously required leaving it.

Part two, the agent that keeps her customers, without taking their data

A Customer Service Point earns from repeat business. A customer who opens an account and never returns is worth almost nothing. Yet the agent has no system for following up. She remembers who came in, roughly, and she calls the ones she remembers.

Our second tool is an autonomous assistant that runs on her phone. It notices which customers are due for a deposit, whose insurance premium is falling due, who opened an account and never transacted again, and it prompts her to reach out, drafting the message for her in her language.

The design decision that matters here is that it runs on the device. Customer information is not uploaded to us, is not held on our servers and is not pooled across agents. The processing happens where the data already is. This was not a technical preference. It is how a system of this kind can be operated honestly under the Digital Personal Data Protection Act, 2023, and it is what allows a bank or a state mission to adopt it without inheriting a data liability.

How it reaches women, and why we do not build a cadre

India already has the cadre. The state rural livelihoods missions, working under the National Rural Livelihoods Mission, have placed tens of thousands of women in Business Correspondent roles as BC Sakhis and have built the self help group federations that select, support and hold them accountable. Building a parallel structure alongside that would be wasteful and would not survive the end of a grant.

So we deploy onto what exists. We integrate with the training and reporting systems institutions already run, and we hand over the tools and the training material under open licences so that the programme continues whether or not we are funded to continue it.

Who this is for

• Women serving as Business Correspondent agents, business facilitators and Customer Service Point operators, including BC Sakhis appointed through state rural livelihoods missions
• The women customers those operators serve, who gain a trusted local point of access to banking, payments, insurance and government entitlements
• State rural livelihoods missions, self help group federations, banks and network managers responsible for the performance of a cadre they cannot supervise face to face

Where we work

As of August 2026 the platform is in live use with 78 women running State Bank of India Customer Service Points across 46 districts in nine states. This first deployment runs through the Customer Service Point network of our affiliate, Eko Bharat Ventures, which is disclosed in full here. We are in discussion with state rural livelihoods missions and with other banks on deployments to their own cadres. We list a state as a state of operation only once a deployment is live in it.

The same business activity that this programme helps a woman grow also becomes the evidence for our access to finance work, where it is turned, with her consent, into a creditworthiness signal that can unlock financing she is otherwise refused.

A woman who has a smartphone, a data pack and an account she can log into is counted as digitally included. In practice she may use two applications, may hand the phone to her son when anything unfamiliar appears, may not know that the message asking for her one time password is a fraud, and may never have used the internet to claim an entitlement, compare a price or start an enterprise. Adoption measured as connection overstates adoption measured as capability.

We work on the second kind. Our approach has four commitments.

• Voice before text. Instruction is delivered by speech in the user's own language, because reading fluency cannot be assumed and because a woman who cannot read is not thereby unable to learn.
• Learning at the point of need. Content arrives when a task is being attempted, not in a scheduled session that requires an hour she does not control.
• A trusted local face. Digital confidence transfers through people, so we route learning through the woman agent at the Customer Service Point rather than around her.
• Safety taught as a skill, not a warning. Recognising a fraudulent call, refusing to share a one time password, checking who is asking and knowing where to complain are practical competencies, and we teach them as such.

This programme is deliberately coupled to the last mile work. The woman agent is both the first user of what we build and the channel through which it reaches other women. Her outlet is the place where a woman who has never transacted digitally is most likely to try for the first time, because there is somebody there who will not make her feel foolish.

A woman running a banking outlet, or a women-led collective processing honey, coffee or grain, generates real business data every day. Transactions, inventory, sales, the rhythm of repayment. Almost none of it becomes a credit history. She is turned away for want of collateral and a formal paper trail, while the evidence that would prove her creditworthiness sits unused on a phone. We build the layer that turns that everyday business data into a signal a lender can act on, we keep it in the woman's control, and we keep it open so any lender can use it.

The problem is not risk. It is legibility.

Lenders are not wrong to be cautious. They are working with the wrong information. A thin-file woman entrepreneur looks risky because the formal system cannot see her business, not because her business is weak. Meanwhile the assets that would raise her income the most are exactly the ones she cannot finance, because they cost more than she can pay upfront and she has no collateral to pledge. The result is a woman running a real, growing business who cannot borrow to grow it. We work on three parts of that gap.

Alternative credit assessment

We use the data a woman's business already generates, digital sales, inventory consistency, cataloguing accuracy, transaction records and market readiness signals, to build a credible, portable picture of creditworthiness for women the formal system scores as thin-file or does not score at all. The assessment is built from her own verified activity rather than from a proxy for her postcode, her community or her husband's record, it travels with her, and it is designed to be read by any lender rather than locked to one.

Financing for clean-energy and productive assets

The assets that raise a woman's income are often climate-smart assets. Decentralised renewable energy equipment, meaning solar-powered processing units, cold storage and efficient manufacturing tools, reduces her costs, extends her working hours and cuts emissions at the same time. We build the data and eligibility layer that lets these assets be financed against a woman's demonstrated business performance, so that clean-energy equipment becomes reachable because her creditworthiness is finally legible, rather than only when a subsidy happens to arrive. Climate is the application. Data about the woman borrower is the foundation that makes it work.

Risk-sharing and blended finance

A good signal is still not enough if the first loss sits entirely with the lender. We help structure the risk-sharing and blended finance arrangements that make lending to these collectives viable: first-loss guarantees, interest subvention, and growth-linked eligibility that ties a larger facility to demonstrated readiness. We design these in the open, so that a structure proven once can be reused by others rather than rebuilt each time.

What we are not, in this programme

We are not a lender, a non-banking financial company or a guarantor. We do not hold the capital, we do not carry the loan book and we do not take a spread on anyone's borrowing. We build the open data and creditworthiness layer, and we convene the equipment manufacturers, the asset financiers, the banks and the blended finance partners who provide the machines and the money.

Data, consent and ownership

The signal is built from the woman's own business data, with her consent, under the Digital Personal Data Protection Act, 2023. Where the assessment can run on data that already lives on the operator's device, it is processed there. The woman can see what is held about her, can carry her record to another lender, and is never scored on data she did not agree to share. A creditworthiness layer that a woman cannot see or control would reproduce the exclusion it is meant to remove. This work shares the same open codebase and the same on-device design as our open technology and responsible AI programmes.

Who this is for

• Women operators and women-led collectives, including self help group enterprises and producer collectives, who run a real business the formal system cannot yet see
• Non-banking financial companies and banks with asset finance and renewable energy portfolios looking for a credible way to lend to thin-file women borrowers
• Manufacturers of productive-use renewable energy equipment seeking aggregated, creditworthy rural demand
• Blended finance facilities, philanthropies and development funders willing to hold first-loss or subvention positions that unlock commercial capital
• State rural livelihoods missions building value chains in sectors such as honey, coffee, grain and crafts

If you finance assets, manufacture them, or carry risk capital that could sit behind them, write to suresh@swandigitalfoundation.in.

Our commitment

Everything the Foundation builds is released as a digital public good. Any bank, state mission, network manager, non profit or company may take the code, run it, adapt it, deploy it to its own users and improve it, without paying us and without asking our permission. We do not operate a paid tier, we do not hold back a proprietary component and we do not grant any organisation an exclusive right to anything we produce.

We are making this a constitutional commitment rather than a policy preference. The revised objects of the Foundation, now being filed, require technology built in furtherance of those objects to be made broadly and equitably available as open source digital public goods under recognised open licences on a royalty free and non exclusive basis, and the revised Articles reserve any decision on the licensing or disposition of intellectual property to a meeting at which independent directors are in the majority.

How we license

The stack

We build on open foundations, chosen deliberately and stated openly.

• The language models are open weight models, principally Google Gemma 3, selected because the licence carries no user threshold clause, permits commercial and non commercial redistribution, and performs adequately in Indian languages at a size that runs on modest hardware. We evaluate other open weight models, including Qwen and Nemotron, for specific tasks.
• Agent orchestration uses LangGraph, which gives explicit, auditable control over what an autonomous agent is permitted to do and where a human must intervene.
• The re-engagement agent executes on the operator's device, so that customer data is processed where it already resides.
• Voice and Indian language interfaces are built for entry level Android handsets on intermittent connectivity, because that is the equipment and the network our users actually have.

Ownership and where the code will live

Copyright in TARA is being assigned to the Foundation by deed from Eko Bharat Ventures Private Limited, which built the initial system. On execution of the deed the Foundation will own the platform outright, the core will carry the AGPL 3.0 licence file, and the repositories will be published with contribution guidance, a code of conduct and a security disclosure process. This page will link them the day they are live. Until then we do not describe the code as public.

Global Access

Where our work is funded by the Gates Foundation or by funders applying comparable terms, we commit to Global Access. Funded technology and the knowledge gained from it will be promptly and broadly disseminated, and will be made available at an affordable price to people most in need in developing countries. Our open licensing arrangement is designed to satisfy that undertaking by default rather than by negotiation, and the revised Articles prevent the Foundation from granting an exclusive right that would defeat it.

Read also: Responsible artificial intelligence and data protection and Our relationship with Eko Bharat Ventures.

We build artificial intelligence for users who have limited means of checking what they are told and limited recourse when it is wrong. That asymmetry sets the standard we hold ourselves to.

What the system will not do

• It does not give individual financial advice, and it does not recommend a specific credit, insurance or investment product to a specific person.
• It does not execute a financial transaction autonomously. Every transaction is initiated and completed by the operator.
• It does not send a message to a customer on its own authority. It drafts, and the operator sends.
• It does not upload customer personal data to our servers.

Accuracy and oversight

• Answers on regulated matters are grounded in a curated body of bank and regulator issued material rather than generated freely, and the source is shown.
• Defined categories of question, including grievances, suspected fraud, account freezes and anything where an incorrect answer would cause financial loss, are escalated to a human channel rather than answered.
• Outputs are sampled and reviewed on a published cadence, with error rates recorded and reported.
• Operators can flag a wrong answer in one action, and flagged answers enter the review queue.

Bias and language

• We evaluate performance separately by language, by dialect and by the user's literacy level, because an aggregate accuracy figure conceals exactly the failure we care about.
• We test for advice that varies with the gender inferred from a name or a voice, and we treat any such variation as a defect.
• Evaluation instruments are published under an open licence so that others can test our claims and reuse the instruments.

Data protection

We build to comply with the Digital Personal Data Protection Act, 2023. Personal data is collected only where it is needed for a stated purpose, notice and consent are given in the user's language, retention is limited and stated, and the customer re-engagement agent processes customer data on the operator's own device so that it is never transmitted to us at all. Our data protection and responsible technology policy is being adopted by the board and will be published on the policies page.

There is a great deal of enthusiasm about artificial intelligence for development and comparatively little evidence. We would like to change the ratio for the specific question we work on, which is whether a coaching and customer acquisition system delivered by phone raises the earnings, retention and confidence of women agents, and whether it increases digital adoption among the women they serve.

How we measure

• Operator earnings, transaction volume and value, product mix and active months, drawn from transaction systems rather than from self report.
• Operator retention and dropout, measured against the cadre baseline.
• Customer adoption, being the number and share of women customers transacting digitally, first time users, and repeat use.
• Confidence, agency and time use, measured through instruments administered by an independent research partner.
• Cost per operator supported and cost per additional woman brought into digital use, published so that others can compare.

How we evaluate

Our evaluation design is a stepped wedge study in which the platform is introduced to blocks of operators in a randomised sequence, so that operators who have not yet received it act as the comparison group for those who have. This provides a credible counterfactual without withholding the intervention from anybody for the duration of the study, which matters when the participants are women whose livelihoods are at stake.

An independent evaluation partner is being appointed, and we intend to align the study with the AI Evidence Alliance for Social Impact so that the findings are comparable with other work in the field. We will name the partner here once an engagement letter is signed.

What we publish

We publish our evaluation protocol before the results, our results whether or not they are favourable, our measurement instruments under an open licence, and our anonymised data where it can be released without risk to any individual. Where a programme does not work we say so and we say what we changed.

The Foundation is small by design. Everything we build is intended to be operated by institutions that already reach the women we serve. Four kinds of partner matter to us.

State rural livelihoods missions and self help group federations

The missions and their federations have selected, trained and placed the largest cadre of women banking agents in the world. We bring tools that raise the earnings and survival of that cadre, deployed through the mission's own structures, integrated with its reporting, and handed over under open licences so the mission owns what it runs.

Banks and Business Correspondent network managers

Agent attrition is expensive and women agents are the hardest to retain. We offer an evidence based way to raise agent productivity and retention that a bank can adopt without a licence fee and without transferring customer data to a third party.

Technology partners and open source contributors

Our design decisions are documented and our repositories will be public. Organisations working on Indian language voice, on device inference and agent safety will find work in our codebase that overlaps with theirs, and we would rather contribute upstream than fork.

Funders

We work with philanthropies, bilateral and multilateral agencies and corporate social responsibility programmes. We are set up for institutional diligence. Our registrations, board composition, policies and affiliate relationship are published, and our revised governance documents place an independent director majority and an open licensing commitment in the Articles rather than in board minutes that could be changed quietly.

How to start a conversation

Write to suresh@swandigitalfoundation.in with a short description of the cadre or population you work with, the states involved and the outcome you are trying to move. If a deployment is feasible we will say so quickly, and if it is not we will say that too.

SWAN Digital Foundation is a not for profit company registered under Section 8 of the Companies Act, 2013. It has no distributable profit. Its income and property are applied solely to the promotion of its objects, and no portion is paid to its members by way of dividend or bonus.

Because the Foundation was established by the founders of a for profit company and continues to contract that company for engineering services, we have not relied on assurance. We are placing the controls in the Articles of Association, where changing them requires the previous approval of the Registrar of Companies, a special resolution of the members and a filing with the Ministry of Corporate Affairs. The revised Articles have been drafted and are being taken through that process; the controls below describe what they require.

The controls

The board

The revised Articles require a board on which independent directors hold the majority and chair it. That board is being constituted. The current directors are named below, and independent directors are being identified and appointed. Until the independent majority is in place, the board does not approve a related party transaction or any other reserved matter, as the Articles require. Each independent director will be listed here on appointment, with independence status, date of appointment and a short biography.

Abhishek Sinha, Director. Co-founder of Eko Bharat Ventures Private Limited. Not independent, by reason of that interest, and recused from board decisions concerning the relationship between the Foundation and Eko Bharat Ventures Private Limited.

Abhinav Sinha, Director. Co-founder of Eko Bharat Ventures Private Limited. Not independent, by reason of that interest, and recused from board decisions concerning the relationship between the Foundation and Eko Bharat Ventures Private Limited.

Audit and financial control

The Foundation is audited annually by a statutory auditor appointed by the members. Grant funds are held in dedicated accounts, are accounted for separately by project, and are reported to funders on the schedule each agreement requires. Audited financial statements are published on our transparency page.

Read also: Our relationship with Eko Bharat Ventures and Policies.

The relationship

SWAN Digital Foundation and Eko Bharat Ventures Private Limited share their founders. Abhishek Sinha and Abhinav Sinha, co-founders of Eko Bharat Ventures Private Limited, a fintech company incorporated in 2007 that operates as a Business Correspondent to State Bank of India and serves a network of Customer Service Point operators among other lines of business, are the current directors of the Foundation.

That shared origin is the reason the Foundation understands this problem. It is also a risk to the Foundation's independence, and we treat it as one.

The Foundation's first live deployment of TARA runs with women operators in Eko's State Bank of India Customer Service Point network. We say so plainly, because it is the fact, and because the controls below exist precisely so that this starting point does not become an exclusive advantage.

What we are doing about it

1. Ownership. Copyright in TARA is being assigned to the Foundation by deed. On execution the Foundation will own it outright. Eko Bharat Ventures Private Limited will hold no ownership interest, no co-ownership interest and no control over it, and no exclusive right of any kind in it.
2. Licensing. The platform core will be licensed to the public under the GNU Affero General Public License version 3. Eko will have the same rights in it as any other organisation in the world, which is to say the rights the licence grants to everybody, and no more.
3. Board control. Independent directors will hold the majority on the Foundation's board and chair it. Any decision concerning the Eko relationship, or concerning the licensing or disposition of the Foundation's intellectual property, will require independent directors to be in the majority of those present and voting.
4. Recusal. The interested director does not count in the quorum, does not participate in the deliberation beyond answering questions of fact, and does not vote on any matter in which he is interested.
5. A cap, disclosed. Eko is engaged as an engineering services vendor on arm's length terms. Aggregate annual payments to Eko are capped by board resolution and disclosed in the annual financial statements. The cap and the actual payments for each financial year will be stated here from the first year in which grant funded engineering work is contracted.
6. Ring fencing. Grant funds are held in dedicated accounts and applied only to the funded project. No grant funds are applied to Eko's commercial products, and no funded work confers exclusive advantage on Eko.
7. No revenue share. The Foundation receives no share of Eko's transaction revenue and pays Eko no share of grant revenue beyond contracted engineering services delivered against invoices.

Why we contract Eko at all

Because the engineering team that built the underlying systems, that understands the Business Correspondent transaction environment and that can deploy to entry level devices in the field is at Eko. Rebuilding that capability inside the Foundation would cost donor money to reproduce something that already exists. We think the honest answer is to buy the service at arm's length, cap it, disclose it, and let independent directors approve it, rather than to construct an arrangement that looks cleaner and delivers less.

What a funder can ask for

We will provide, on request, the deed of assignment once executed, the board resolution fixing the vendor cap, the register of interests, the minutes recording recusal on relevant matters, and the notes to the financial statements disclosing related party transactions. Write to suresh@swandigitalfoundation.in.

Legal and registration details

Documents

The following are available to funders and their auditors on request, and will be published on this page as each is finalised.

• Certificate of Incorporation and the fresh certificate recording the change of name
• Memorandum and Articles of Association, as currently in force, and the revised versions once filed
• Audited financial statements for the last three financial years
• Annual reports
• Board composition, with independence status for each director
• Policies: conflict of interest, related party transactions, intellectual property and open licensing, data protection and responsible technology, prevention of sexual harassment at the workplace, child protection and safeguarding, whistleblower, and anti bribery and anti corruption
• Related party transaction disclosures

Funding

We will publish the name of each institutional funder, the purpose of the grant, the amount and the period, from the first institutional grant received. We do not accept funding that is conditioned on an exclusive right to anything we build, or that would prevent us from licensing our work openly.

Grievances and safeguarding

Concerns about the conduct of any person associated with the Foundation, including concerns about the safety of a woman or a child in a programme we support, may be raised at pooja@swandigitalfoundation.in or by post to the registered office marked for the attention of the Director. Complaints under the Sexual Harassment of Women at Workplace (Prevention, Prohibition and Redressal) Act, 2013 may be made to the Internal Committee at the same address. We will acknowledge within five working days.

The revised Articles require the board to adopt, maintain and periodically review the policies below and to make them available to donors and their auditors on request. Each will be published here on adoption.

• Conflict of interest policy
• Related party transactions policy
• Intellectual property and open licensing policy
• Data protection and responsible technology policy
• Prevention of sexual harassment at the workplace policy
• Child protection and safeguarding policy
• Whistleblower policy
• Anti bribery and anti corruption policy

Requests for current drafts may be sent to suresh@swandigitalfoundation.in.

Work with us

We are a small team building software for users who will never be in the room when we build it. That places an unusual demand on the people we hire, which is that they spend time at a Customer Service Point in a village before they write a line of code for one.

We hire engineers, designers, programme managers, field researchers and evaluation specialists. We also welcome contributions to our open repositories from people we will never employ, and we treat a good pull request as seriously as a good application.

There are no open roles at present. Speculative applications may be sent to pooja@swandigitalfoundation.in.

This website is a static site. It does not set cookies, it does not run analytics or advertising scripts, and it does not collect personal data from visitors. Fonts are loaded from Google Fonts, which may record the network request in accordance with Google's own privacy policy.

If you write to us at any of the addresses on this site, we keep your message for as long as needed to respond and to keep a record of the correspondence, and we do not share it beyond the Foundation except where the law requires. You may ask us to delete correspondence by writing to the address you used.

Personal data processed within the Foundation's programmes is governed by the data protection and responsible technology policy described on the responsible AI page, and by the Digital Personal Data Protection Act, 2023.