FiBi

Project Overview

FiBi is a product within the SeamlessVisa platform focused on the interview stage of the visa process. After an application is approved and an interview is scheduled at the embassy or consulate, the applicant still has to show up and perform well under questioning — often under significant pressure and with high personal stakes. FiBi exists to close that preparation gap.

It is an AI interview coach that conducts mock visa interviews. It knows the applicant — either from their existing SeamlessVisa profile or from a PDF of their application that they upload — and uses that data to generate questions that are specific to their visa category, their situation, and the kinds of questions a consular officer would actually ask. During the session, it evaluates responses and gives feedback in real time.

I was part of the front-end engineering team at SeamlessVisa. FiBi currently focuses on US visa interviews.

Session Modes

FiBi supports multiple input and response configurations so applicants can practice in the format that suits them:

• Input modes: Voice only (hands-free), text only (silent mode), or voice and text together for maximum flexibility
• Response modes: Text only (silent feedback) or text and voice combined (spoken responses from FiBi alongside written feedback)

The combination of modes means someone practicing alone late at night can run a fully silent text session, while someone who wants to simulate the spoken nature of an actual consular interview can use voice in and voice out.

The Problem

A visa approval does not end the process. For many visa categories — particularly US visas — the applicant must attend an in-person interview at the embassy or consulate before the visa is physically issued. The interview is a genuine gate: an officer asks questions, evaluates the applicant's answers and demeanour, and makes a decision. Applicants can be denied at this stage regardless of how strong their paperwork was.

Preparing for this interview is difficult because:

• The questions are not published in advance. Officers draw from the applicant's specific application — travel history, employment, financial situation, purpose of travel, connections to the home country — so generic preparation is less useful than application-specific practice.
• The format is verbal and in-person, under time pressure and with an authority figure. Reading about what to expect and actually being asked questions are very different experiences. Applicants who haven't rehearsed verbal responses often stumble on things they know the answer to simply because they've never said it out loud.
• Human practice partners are either unavailable or uninformed. Friends and family can role-play, but they don't know what a consular officer actually looks for, what follow-up questions typically come, or how to give useful feedback on an answer.

Application-Aware Question Generation

The defining feature of FiBi compared to a generic interview prep tool is that it knows who it's interviewing. Before the session starts, FiBi has access to the applicant's data — either pulled from their SeamlessVisa profile if they've used the platform's form engine, or ingested from an uploaded PDF of their application.

That data feeds directly into question generation. FiBi asks about the applicant's specific travel history, their stated purpose for travel, their employment and financial circumstances, their ties to their home country — the same things a consular officer would probe based on the same application. The questions are not generic; they are drawn from the applicant's own situation.

Real-Time Feedback

After each response, FiBi evaluates what the applicant said and delivers feedback immediately. The feedback is not just a score — it tells the applicant specifically:

• What they said that was strong and why it would read well to an officer
• What they said that was weak, ambiguous, or potentially concerning — and how to reframe or address it
• What a better answer to that question would look like
• What follow-up questions their answer might invite

The goal is not just to tell applicants whether they answered correctly — it's to help them understand how a consular officer reads their responses, so they can adjust their approach before the real thing.

Session Dashboard

During a session, the dashboard surfaces:

• A live timer with pause and resume controls
• Mode toggle between text and voice mid-session if needed
• A stage progress indicator showing where in the interview the session currently is
• A readiness score that updates as the session progresses, reflecting overall performance so far
• Confidence indicators showing how responses are being read by the AI

After the session ends, the full transcript is saved and available for review. Applicants can go back through their answers, re-read the feedback, and track how their performance changes across multiple sessions.

Outcomes

• FiBi is live at seamlessvisa.com/fibi.
• The application-aware approach — drawing questions from the applicant's own data rather than a generic question bank — makes FiBi more specific and useful than a static interview prep guide. Applicants practice the actual questions their application is likely to generate, not a generic set.
• The combination of text and voice modes means FiBi covers both the content of answers (what to say) and the delivery (how to say it under the pressure of a verbal interaction).

Key Learnings

• Context-aware tools require solid data pipelines: FiBi's quality depends entirely on the quality and completeness of the applicant data it receives. When a PDF upload is incomplete or poorly structured, the questions it generates are less targeted. The front-end data ingestion and parsing flow matters as much as the AI layer.
• Feedback that explains the why is more useful than a score: Early feedback design leaned toward rating responses numerically. User behaviour showed that applicants engaged more deeply with textual feedback that explained the reasoning — what a consular officer would think and why — than with a number they didn't know how to act on.
• Voice mode introduces UX problems that text mode doesn't: Turn-taking, mic permissions, handling silence, detecting when an answer is complete — these are problems that don't exist in text mode and each requires specific front-end handling. They're not hard problems individually, but they compound when something goes wrong mid-session.