VivaBoard
- Founder
- 2025 - present
The problem
A BCS candidate in Bangladesh can find a hundred books of model questions and not one honest rehearsal. The viva is the part of the exam that decides careers, and it is the only part nobody can practise alone. The same is true for bank recruitment panels, university admission interviews and the first job after graduation.
What candidates lack is not information. It is a room with someone in it who will ask a second question - the one that follows from what you actually just said, rather than the one printed next in the list.
The interview
VivaBoard AI is a bilingual mobile app that runs a realistic AI-led mock interview. The examiner speaks its questions aloud. The candidate answers with their voice and sees a live transcription as they speak. The system generates follow-up and counter questions from the content of the answer, not from a fixed script, which is the part that makes it feel like a viva rather than a quiz.
The report afterwards covers dimension-wise scores, strengths and weaknesses, filler-word analysis, confidence and body-language signals derived from on-device face analysis, and the full transcript, reviewable line by line.
The preparation ecosystem around it
A mock interview is only useful if there is somewhere to go after a bad one. Around the interview engine sits a learning centre organised by sector and chapter, with video and audio classes, mock tests, quizzes and searchable question banks; exam preparation questions and answers; blogs and daily suggestions; a live job circular feed with saves and alerts; preparation streaks and reminders; a leaderboard; and in-app support.
It is monetised through subscription packages with coupons and referral codes, alongside an ad-supported free tier, because a candidate who cannot pay is exactly the candidate the product exists for.
VivaBoard Hire
The same engine, pointed at employers. Companies create jobs and interview campaigns, configure their own questions and scoring rubrics, invite candidates to asynchronous AI-conducted interviews, and receive transcripts, competency scores, integrity signals and a ranked pipeline.
It is positioned as decision support for recruiters. It ranks and evidences; it does not reject.
The decision
Almost nothing is hardcoded
Pricing, credit costs, scoring dimensions, the number of questions in a session, ad placements, the weighting of each report metric - all of it lives in the admin panel as configuration rather than in the application as code. A product that has not launched does not yet know what its scoring rubric should be, and the cost of finding out should be an afternoon, not a release cycle.
The second decision follows from the first. Hire returns evidence, not verdicts. An AI that scores a candidate and an AI that filters a candidate out are the same model with different consequences, and the second one is a system I do not want to have built. Recruiters get the transcript, the competency scores and the integrity signals, and they make the call.
The speech layer was the hard part. English speech-to-text is a solved commodity; Bangla is not, and a mock interview that mishears the answer is worse than no mock interview, because it teaches the candidate the wrong lesson. Sarvam handles Bangla speech-to-text and text-to-speech, OpenAI handles reasoning and evaluation, and the split exists because no single vendor was good enough at both ends.
Stack
Expo, React Native
Firebase, Cloud Functions (Mumbai region)
Sarvam - Bangla STT and TTS
OpenAI
EPS
Status
Pre-launch, in closed testing. Bangladesh first; the architecture is built for elsewhere.
Next
Local discovery for Bangladesh where you cannot leave a review without uploading the receipt - proof first, opinion second.