InterviewKo: AI mock interviews that coach Filipino jobseekers on what they say, how they say it, and how confident they sound

Problem Statement

Hundreds of thousands of Filipinos graduate from college every year, and BPO firms and overseas employers hire at large scale. Recruiters often say that many applicants fail at the interview and communication stage, not because they lack ability. Applicants from provinces and state universities rarely have access to mock interviews, English coaches, or industry mentors, and anxiety and unfamiliarity with interview structure hurt them. BPO versant-style assessments and OFW employer interviews are high-stakes. Private review centers charge fees, while school career offices and PESOs have few staff to coach applicants one-on-one.

Proposed Solution

InterviewKo is a mobile-first web app. The user picks a target, such as “BPO voice account – telco,” “Caregiver – Japan,” “Nurse – UK,” “Seafarer,” or “Fresh grad – bank teller,” or pastes a real job post. An AI interviewer with a voice and optional avatar asks realistic questions and follow-ups, including behavioral STAR questions, customer-call role-plays, and situational questions. After each answer the user gets: content feedback (relevance, structure, specificity, and a rewritten model answer), fluency feedback (pace, filler words, grammar, and pronunciation of difficult words, with no accent shaming), and confidence signals (pauses, volume steadiness, and optionally eye contact and posture from the webcam). A progress tracker shows improvement across sessions. Filipino-language explanations help users understand the coaching, and a “kaba mode” eases anxious users in with warm-up questions.

Target Market

• Primary users: fresh graduates, BPO applicants, and OFW applicants (nurses, caregivers, seafarers, skilled workers), together several hundred thousand to over a million jobseekers each year.
• B2B buyers: BPO recruiters and IBPAP (pre-screening and training funnels), licensed recruitment agencies, and universities’ career offices.
• B2G buyers: PESOs, the DMW pre-employment orientation ecosystem, TESDA, and CHED career-readiness programs.

How AI Is Used

• Speech-to-text with timestamps: captures exactly what was said and how, which fluency analysis needs.
• LLM interviewer and evaluator: generates role-specific questions and follow-ups and scores answers against rubrics (STAR, relevance). Fixed question banks can’t follow up or critique.
• Pronunciation and fluency scoring: measures words per minute, fillers (“ano,” “uhm”), and phoneme-level clarity using alignment models.
• Prosody and computer vision (optional): detects confidence cues such as long pauses, pitch variance, and gaze, giving feedback that human-free practice otherwise lacks.
• TTS interviewer voice: makes practice feel real and builds composure.

PLDT / Smart / ePLDT / Rokid Angle

• PLDT Enterprise: a value-add for BPO clients’ recruitment pipelines, bundled with contact-center connectivity and cloud services.
• ePLDT cloud: low-latency speech processing hosted locally, sold as SaaS to universities and LGUs.
• Smart: data promos for jobseekers and SMS interview-practice reminders.
• Rokid (stretch): AR glasses simulate a panel interview room so OFW applicants can practice in an immersive setting.

Demo Scenario (3 minutes for judges)

1. “Paolo,” a fresh IT graduate from a state university in Leyte, has a BPO final interview tomorrow and is nervous.
2. Live, he selects “BPO Voice – Tech Support.” The AI interviewer asks, “Tell me about a time you solved a difficult problem.”
3. Paolo answers on stage. Within seconds the feedback appears: content 6/10 (no result stated), 7 filler words, “pronounce ‘specifically’ like this,” and a long pause at 0:42.
4. The AI shows a stronger STAR version of his own story.
5. Second attempt: the scores rise visibly on the progress chart.
6. A quick role-play follows with an irate AI customer.
7. Close: “Hindi ka kulang sa galing. Kulang ka lang sa practice.” (You’re not short on skill. You’re short on practice.)

30-Hour Build Plan

Suggested stack: Next.js, WebRTC or MediaRecorder, FastAPI, Whisper (word timestamps), Gemini or GPT-class LLM, Azure Speech Pronunciation Assessment or wav2vec2 forced alignment, MediaPipe Face Mesh (optional), ElevenLabs or Google TTS. Public data / APIs: sample public job posts from PhilJobNet and the DMW job order listings, PSA OpenSTAT labor statistics, Hugging Face speech datasets (e.g., L2-ARCTIC and FLEURS) for evaluation, and public BPO competency frameworks.

Risks & Privacy Notes

• Accent bias: score intelligibility, not “native-likeness.” Test across Philippine accents and keep the feedback encouraging.
• Over-coaching or scripted answers: model answers reuse the user’s own experience. Make clear it is practice, not a hiring decision tool.
• Data Privacy Act 2012: voice and video are personal data. Get explicit consent, process audio and video then discard them by default, store only scores, host in PH, and share nothing with employers without opt-in.
• Clause 24 compliance: the demo uses team members’ own voices with consent and fictional job posts. AI use is disclosed.

AI Use Disclosure (draft)

We use Whisper for transcription, a Gemini or GPT-class LLM for the interviewer and feedback, a pronunciation assessment model for fluency scoring, and a TTS engine for the interviewer voice. Coding assistants helped with UI boilerplate, and an LLM helped draft this proposal. Team members reviewed rubric outputs for fairness and accuracy. No real applicant data was used.

Rubric Scoring