
The problem. Entry-level hiring relies on unstructured, unrecorded phone screens prone to bias, while written tests are easily faked. What we built. Interviewa AI is a voice-first workspace. Recruiters set up jobs with custom rubrics and share secure links. Candidates conduct real-time AI voice interviews with zero typing or scheduling required. How the voice works. The browser streams 24 kHz audio via WebSockets to AssemblyAI using FastAPI-minted tokens. Mid-conversation voice tools (score, pause, rephrase, switch_language, request_human, finish) allow candidates to pause, rephrase, switch English/Kiswahili, or request a human. AI collects evidence, humans decide. The AI never hires, rejects, or ranks; it only gathers competency evidence. Hiring teams evaluate transcripts, evidence, and follow-ups on a review dashboard with audit trails. Interviewer Simulator. Recruiters practice voice interviews on five AI candidate personas (e.g., nervous junior, overconfident candidate). Also included. Multi-tenant isolation (company_id), RBAC, versioned configs, PDF resume screening, bilingual support, HTTP-only JWTs, demo data, and pytest suite. Tech stack. Python, FastAPI, vanilla JS, Tailwind, SQLite, and SQLAlchemy—no React, Node build steps, or external auth. Runs instantly via pip install -r requirements.txt and uvicorn app.main:app. What's next. Postgres/object storage, enterprise SSO/MFA, automated retention, and ATS webhooks.
30 Sep 2026