The HR technology company Hipo AI formally opened its artificial-intelligence-run real-time interview platform on Sunday, making it available to every employer and job seeker in Bangladesh. Before the launch it ran a pilot in which three employers and more than 14,000 candidates completed interviews through the system.
What it does
A candidate creates an account, finds an opening and sits a real-time interview conducted by the model rather than by a person. The employer gets a structured assessment instead of a stack of CVs, and can put more candidates through the same process in less time. There is a mobile app as well as the website at hipo.com.bd, which matters in a market where most people reach the internet through a phone.
The company frames the problem as wasted time on both sides: screening takes employers weeks, and candidates rarely learn why they were dropped.
The claim being made
Co-founder and chief executive Md Ziaul Parvez Chowdhury put the case in terms of fairness rather than efficiency. "Our journey began from the thought of making the recruitment process equal for everyone," he said, arguing that unemployment cannot be solved while job seekers do not get a fair chance, and that conventional HR technology cannot deliver one. The stated goal is interviews in every language including the mother tongue, with particular weight on the Bangla interview experience — which the company says is the first of its kind in Bangladesh.
Why the Bangla interview is the substantial part
AI interview tools are not new. Firms in the United States and India have sold them for years, and almost all of them operate in English.
In Bangladesh that detail decides who passes. English fluency here tracks schooling, and schooling tracks money and geography. A graduate from a Dhaka English-medium school and a graduate from a district college may know the same amount about accounting, inventory or customer service, and sound completely different describing it. An English-only screen measures the second thing while appearing to measure the first, and it does so before any human sees the candidate.
Conducting the interview in Bangla removes that filter. It does not make the assessment fair on its own, but it stops one specific unfairness that English-only tooling imports by default — and it is the kind of adaptation that has to be built locally, because no foreign vendor has a commercial reason to do it.
What 14,000 interviews establish, and what they do not
The pilot number is real and the platform evidently works at that volume. Completion is not accuracy, though. Fourteen thousand finished interviews show the pipe carries traffic; they say nothing about whether the model's judgement of a candidate matches what a good human interviewer would have concluded, and nothing about whether the three participating employers found the recommendations useful.
The deeper question applies to every automated screen. A human interviewer having a bad morning affects one candidate in one room. A model that has learned to prefer a particular way of speaking applies that preference to every candidate, identically, invisibly, and at a scale no complaint process will catch. Bias in a person is a problem; bias in a screening system is a policy.
That is testable, and the test is publication. Pass rates broken down by district, by university, by gender and by whether the candidate chose the Bangla or the English track would settle in one table what any amount of description cannot. A company making fairness its central claim is the one best placed to publish them.
If you are about to sit one
- Choose the Bangla track if your Bangla is stronger. It exists so you do not have to compete on a second language, and using it is not a mark against you.
- Test the microphone before you start, in the room you will actually sit in. A real-time system is transcribing you; a bad connection reads as a bad answer.
- Answer the question that was asked. Structured interviews score against the question, so a strong answer to a different question scores nothing.
- Give one concrete example per answer — what you did, at which organisation, and what changed because of it. Specifics survive automated scoring better than adjectives.
- Keep a note of what you were asked. If a decision looks wrong, the questions are the only record you will have.
On whether the schools feeding this pipeline are ready for any of it, see the Seldon report on AI in Bangladeshi classrooms. On the same argument one level up the labour market, see the BPO summit and the advice to move up the value chain.
