IndiaPublic sector2016 – ongoing

Telangana state government

Samagra Vedika welfare eligibility algorithm

An Al Jazeera investigation found Telangana's welfare algorithm wrongly denied food cards to at least 7.5% of cases it could check

In progressPress reporting

Problem

The state wanted to remove ineligible and duplicate beneficiaries from food and other welfare schemes. Officials had data spread across dozens of departmental databases.

Approach

Posidex Technologies built Samagra Vedika, which uses machine learning to match people across more than 30 government databases and build one profile each. It began in 2016 as a police tool, was piloted in the food security scheme that year and covered most state welfare schemes by 2018. Officials must consult it before approving applications and in practice usually defer to it.

Results

1.86 millionexisting food security cards cancelled in 2014-2019, reportedly without notice (plus 142,086 rejected new applications)Press reporting[1][2]

7.5%+share of rejected cards found wrongly denied in the processed part of a court-ordered re-verification, Al Jazeera's estimatePress reporting[1]

<5%tagging error rate the state government claimedCompany's own claim[1]

Takeaway

Officials usually followed the algorithm, and a court-ordered review found wrongful denials despite the state claiming tagging errors below 5%. The review covered only some cancelled cards, and the state refused to release source code.

Sources

How an algorithm denied food to thousands of poor in India's Telangana
Al Jazeera24 Jan 2024News article
Algorithmic Welfare Exclusion and the Right to Food in India: Lessons from Samagra Vedika
Oxford Human Rights Hub19 Feb 2026Research paper or report

Source summaries

[1]Al Jazeera24 Jan 2024

Al Jazeera investigated the Posidex-built system and reported that over 1.86 million food cards were cancelled and 142,086 applications rejected in 2014-2019. Using data from a Supreme Court-ordered re-verification, the reporters estimated at least 7.5% of rejected cards were wrongly denied. Officials defended the system, said errors were under 5% and refused requests for source code.

[2]Oxford Human Rights Hub19 Feb 2026

Oxford Human Rights Hub described the system as an entity-resolution tool drawing on over 30 databases and reported the 1.86 million cancellations and 142,086 rejections. It says neither the vendor nor the state has published source code or accuracy data, and that officials in practice defer to the output. It also cited a 2024 Amnesty International investigation raising human rights concerns.

Full details are in the original sources.

Added 25 Oct 2026 · Updated 25 Oct 2026